drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/compat.h>
  10 #include <linux/delay.h>
  11 #include <linux/errno.h>
  12 #include <linux/hdreg.h>
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/slab.h>
  18 #include <linux/types.h>
  19 #include <linux/pr.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/nvme_ioctl.h>
  22 #include <linux/pm_qos.h>
  23 #include <asm/unaligned.h>
  24
  25 #include "nvme.h"
  26 #include "fabrics.h"
  27
  28 #define CREATE_TRACE_POINTS
  29 #include "trace.h"
  30
  31 #define NVME_MINORS             (1U << MINORBITS)
  32
  33 unsigned int admin_timeout = 60;
  34 module_param(admin_timeout, uint, 0644);
  35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  36 EXPORT_SYMBOL_GPL(admin_timeout);
  37
  38 unsigned int nvme_io_timeout = 30;
  39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  42
  43 static unsigned char shutdown_timeout = 5;
  44 module_param(shutdown_timeout, byte, 0644);
  45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  46
  47 static u8 nvme_max_retries = 5;
  48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  50
  51 static unsigned long default_ps_max_latency_us = 100000;
  52 module_param(default_ps_max_latency_us, ulong, 0644);
  53 MODULE_PARM_DESC(default_ps_max_latency_us,
  54                  "max power saving latency for new devices; use PM QOS to change per device");
  55
  56 static bool force_apst;
  57 module_param(force_apst, bool, 0644);
  58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  59
  60 static bool streams;
  61 module_param(streams, bool, 0644);
  62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  63
  64 /*
  65  * nvme_wq - hosts nvme related works that are not reset or delete
  66  * nvme_reset_wq - hosts nvme reset works
  67  * nvme_delete_wq - hosts nvme delete works
  68  *
  69  * nvme_wq will host works such as scan, aen handling, fw activation,
  70  * keep-alive, periodic reconnects etc. nvme_reset_wq
  71  * runs reset works which also flush works hosted on nvme_wq for
  72  * serialization purposes. nvme_delete_wq host controller deletion
  73  * works which flush reset works for serialization.
  74  */
  75 struct workqueue_struct *nvme_wq;
  76 EXPORT_SYMBOL_GPL(nvme_wq);
  77
  78 struct workqueue_struct *nvme_reset_wq;
  79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  80
  81 struct workqueue_struct *nvme_delete_wq;
  82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  83
  84 static LIST_HEAD(nvme_subsystems);
  85 static DEFINE_MUTEX(nvme_subsystems_lock);
  86
  87 static DEFINE_IDA(nvme_instance_ida);
  88 static dev_t nvme_ctrl_base_chr_devt;
  89 static struct class *nvme_class;
  90 static struct class *nvme_subsys_class;
  91
  92 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  94                                            unsigned nsid);
  95
  96 /*
  97  * Prepare a queue for teardown.
  98  *
  99  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 100  * the capacity to 0 after that to avoid blocking dispatchers that may be
 101  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 102  * be synced.
 103  */
 104 static void nvme_set_queue_dying(struct nvme_ns *ns)
 105 {
 106         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 107                 return;
 108
 109         blk_set_queue_dying(ns->queue);
 110         blk_mq_unquiesce_queue(ns->queue);
 111
 112         set_capacity_and_notify(ns->disk, 0);
 113 }
 114
 115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 116 {
 117         /*
 118          * Only new queue scan work when admin and IO queues are both alive
 119          */
 120         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 121                 queue_work(nvme_wq, &ctrl->scan_work);
 122 }
 123
 124 /*
 125  * Use this function to proceed with scheduling reset_work for a controller
 126  * that had previously been set to the resetting state. This is intended for
 127  * code paths that can't be interrupted by other reset attempts. A hot removal
 128  * may prevent this from succeeding.
 129  */
 130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 131 {
 132         if (ctrl->state != NVME_CTRL_RESETTING)
 133                 return -EBUSY;
 134         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 135                 return -EBUSY;
 136         return 0;
 137 }
 138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 139
 140 static void nvme_failfast_work(struct work_struct *work)
 141 {
 142         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 143                         struct nvme_ctrl, failfast_work);
 144
 145         if (ctrl->state != NVME_CTRL_CONNECTING)
 146                 return;
 147
 148         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 149         dev_info(ctrl->device, "failfast expired\n");
 150         nvme_kick_requeue_lists(ctrl);
 151 }
 152
 153 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 154 {
 155         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 156                 return;
 157
 158         schedule_delayed_work(&ctrl->failfast_work,
 159                               ctrl->opts->fast_io_fail_tmo * HZ);
 160 }
 161
 162 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 163 {
 164         if (!ctrl->opts)
 165                 return;
 166
 167         cancel_delayed_work_sync(&ctrl->failfast_work);
 168         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 169 }
 170
 171
 172 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 173 {
 174         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 175                 return -EBUSY;
 176         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 177                 return -EBUSY;
 178         return 0;
 179 }
 180 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 181
 182 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 183 {
 184         int ret;
 185
 186         ret = nvme_reset_ctrl(ctrl);
 187         if (!ret) {
 188                 flush_work(&ctrl->reset_work);
 189                 if (ctrl->state != NVME_CTRL_LIVE)
 190                         ret = -ENETRESET;
 191         }
 192
 193         return ret;
 194 }
 195
 196 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 197 {
 198         dev_info(ctrl->device,
 199                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 200
 201         flush_work(&ctrl->reset_work);
 202         nvme_stop_ctrl(ctrl);
 203         nvme_remove_namespaces(ctrl);
 204         ctrl->ops->delete_ctrl(ctrl);
 205         nvme_uninit_ctrl(ctrl);
 206 }
 207
 208 static void nvme_delete_ctrl_work(struct work_struct *work)
 209 {
 210         struct nvme_ctrl *ctrl =
 211                 container_of(work, struct nvme_ctrl, delete_work);
 212
 213         nvme_do_delete_ctrl(ctrl);
 214 }
 215
 216 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 217 {
 218         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 219                 return -EBUSY;
 220         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 221                 return -EBUSY;
 222         return 0;
 223 }
 224 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 225
 226 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 227 {
 228         /*
 229          * Keep a reference until nvme_do_delete_ctrl() complete,
 230          * since ->delete_ctrl can free the controller.
 231          */
 232         nvme_get_ctrl(ctrl);
 233         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 234                 nvme_do_delete_ctrl(ctrl);
 235         nvme_put_ctrl(ctrl);
 236 }
 237
 238 static blk_status_t nvme_error_status(u16 status)
 239 {
 240         switch (status & 0x7ff) {
 241         case NVME_SC_SUCCESS:
 242                 return BLK_STS_OK;
 243         case NVME_SC_CAP_EXCEEDED:
 244                 return BLK_STS_NOSPC;
 245         case NVME_SC_LBA_RANGE:
 246         case NVME_SC_CMD_INTERRUPTED:
 247         case NVME_SC_NS_NOT_READY:
 248                 return BLK_STS_TARGET;
 249         case NVME_SC_BAD_ATTRIBUTES:
 250         case NVME_SC_ONCS_NOT_SUPPORTED:
 251         case NVME_SC_INVALID_OPCODE:
 252         case NVME_SC_INVALID_FIELD:
 253         case NVME_SC_INVALID_NS:
 254                 return BLK_STS_NOTSUPP;
 255         case NVME_SC_WRITE_FAULT:
 256         case NVME_SC_READ_ERROR:
 257         case NVME_SC_UNWRITTEN_BLOCK:
 258         case NVME_SC_ACCESS_DENIED:
 259         case NVME_SC_READ_ONLY:
 260         case NVME_SC_COMPARE_FAILED:
 261                 return BLK_STS_MEDIUM;
 262         case NVME_SC_GUARD_CHECK:
 263         case NVME_SC_APPTAG_CHECK:
 264         case NVME_SC_REFTAG_CHECK:
 265         case NVME_SC_INVALID_PI:
 266                 return BLK_STS_PROTECTION;
 267         case NVME_SC_RESERVATION_CONFLICT:
 268                 return BLK_STS_NEXUS;
 269         case NVME_SC_HOST_PATH_ERROR:
 270                 return BLK_STS_TRANSPORT;
 271         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 272                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 273         case NVME_SC_ZONE_TOO_MANY_OPEN:
 274                 return BLK_STS_ZONE_OPEN_RESOURCE;
 275         default:
 276                 return BLK_STS_IOERR;
 277         }
 278 }
 279
 280 static void nvme_retry_req(struct request *req)
 281 {
 282         unsigned long delay = 0;
 283         u16 crd;
 284
 285         /* The mask and shift result must be <= 3 */
 286         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 287         if (crd)
 288                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 289
 290         nvme_req(req)->retries++;
 291         blk_mq_requeue_request(req, false);
 292         blk_mq_delay_kick_requeue_list(req->q, delay);
 293 }
 294
 295 enum nvme_disposition {
 296         COMPLETE,
 297         RETRY,
 298         FAILOVER,
 299 };
 300
 301 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 302 {
 303         if (likely(nvme_req(req)->status == 0))
 304                 return COMPLETE;
 305
 306         if (blk_noretry_request(req) ||
 307             (nvme_req(req)->status & NVME_SC_DNR) ||
 308             nvme_req(req)->retries >= nvme_max_retries)
 309                 return COMPLETE;
 310
 311         if (req->cmd_flags & REQ_NVME_MPATH) {
 312                 if (nvme_is_path_error(nvme_req(req)->status) ||
 313                     blk_queue_dying(req->q))
 314                         return FAILOVER;
 315         } else {
 316                 if (blk_queue_dying(req->q))
 317                         return COMPLETE;
 318         }
 319
 320         return RETRY;
 321 }
 322
 323 static inline void nvme_end_req(struct request *req)
 324 {
 325         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 326
 327         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 328             req_op(req) == REQ_OP_ZONE_APPEND)
 329                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 330                         le64_to_cpu(nvme_req(req)->result.u64));
 331
 332         nvme_trace_bio_complete(req);
 333         blk_mq_end_request(req, status);
 334 }
 335
 336 void nvme_complete_rq(struct request *req)
 337 {
 338         trace_nvme_complete_rq(req);
 339         nvme_cleanup_cmd(req);
 340
 341         if (nvme_req(req)->ctrl->kas)
 342                 nvme_req(req)->ctrl->comp_seen = true;
 343
 344         switch (nvme_decide_disposition(req)) {
 345         case COMPLETE:
 346                 nvme_end_req(req);
 347                 return;
 348         case RETRY:
 349                 nvme_retry_req(req);
 350                 return;
 351         case FAILOVER:
 352                 nvme_failover_req(req);
 353                 return;
 354         }
 355 }
 356 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 357
 358 /*
 359  * Called to unwind from ->queue_rq on a failed command submission so that the
 360  * multipathing code gets called to potentially failover to another path.
 361  * The caller needs to unwind all transport specific resource allocations and
 362  * must return propagate the return value.
 363  */
 364 blk_status_t nvme_host_path_error(struct request *req)
 365 {
 366         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 367         blk_mq_set_request_complete(req);
 368         nvme_complete_rq(req);
 369         return BLK_STS_OK;
 370 }
 371 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 372
 373 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 374 {
 375         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 376                                 "Cancelling I/O %d", req->tag);
 377
 378         /* don't abort one completed request */
 379         if (blk_mq_request_completed(req))
 380                 return true;
 381
 382         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 383         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 384         blk_mq_complete_request(req);
 385         return true;
 386 }
 387 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 388
 389 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 390 {
 391         if (ctrl->tagset) {
 392                 blk_mq_tagset_busy_iter(ctrl->tagset,
 393                                 nvme_cancel_request, ctrl);
 394                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 395         }
 396 }
 397 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 398
 399 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 400 {
 401         if (ctrl->admin_tagset) {
 402                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 403                                 nvme_cancel_request, ctrl);
 404                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 405         }
 406 }
 407 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 408
 409 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 410                 enum nvme_ctrl_state new_state)
 411 {
 412         enum nvme_ctrl_state old_state;
 413         unsigned long flags;
 414         bool changed = false;
 415
 416         spin_lock_irqsave(&ctrl->lock, flags);
 417
 418         old_state = ctrl->state;
 419         switch (new_state) {
 420         case NVME_CTRL_LIVE:
 421                 switch (old_state) {
 422                 case NVME_CTRL_NEW:
 423                 case NVME_CTRL_RESETTING:
 424                 case NVME_CTRL_CONNECTING:
 425                         changed = true;
 426                         fallthrough;
 427                 default:
 428                         break;
 429                 }
 430                 break;
 431         case NVME_CTRL_RESETTING:
 432                 switch (old_state) {
 433                 case NVME_CTRL_NEW:
 434                 case NVME_CTRL_LIVE:
 435                         changed = true;
 436                         fallthrough;
 437                 default:
 438                         break;
 439                 }
 440                 break;
 441         case NVME_CTRL_CONNECTING:
 442                 switch (old_state) {
 443                 case NVME_CTRL_NEW:
 444                 case NVME_CTRL_RESETTING:
 445                         changed = true;
 446                         fallthrough;
 447                 default:
 448                         break;
 449                 }
 450                 break;
 451         case NVME_CTRL_DELETING:
 452                 switch (old_state) {
 453                 case NVME_CTRL_LIVE:
 454                 case NVME_CTRL_RESETTING:
 455                 case NVME_CTRL_CONNECTING:
 456                         changed = true;
 457                         fallthrough;
 458                 default:
 459                         break;
 460                 }
 461                 break;
 462         case NVME_CTRL_DELETING_NOIO:
 463                 switch (old_state) {
 464                 case NVME_CTRL_DELETING:
 465                 case NVME_CTRL_DEAD:
 466                         changed = true;
 467                         fallthrough;
 468                 default:
 469                         break;
 470                 }
 471                 break;
 472         case NVME_CTRL_DEAD:
 473                 switch (old_state) {
 474                 case NVME_CTRL_DELETING:
 475                         changed = true;
 476                         fallthrough;
 477                 default:
 478                         break;
 479                 }
 480                 break;
 481         default:
 482                 break;
 483         }
 484
 485         if (changed) {
 486                 ctrl->state = new_state;
 487                 wake_up_all(&ctrl->state_wq);
 488         }
 489
 490         spin_unlock_irqrestore(&ctrl->lock, flags);
 491         if (!changed)
 492                 return false;
 493
 494         if (ctrl->state == NVME_CTRL_LIVE) {
 495                 if (old_state == NVME_CTRL_CONNECTING)
 496                         nvme_stop_failfast_work(ctrl);
 497                 nvme_kick_requeue_lists(ctrl);
 498         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 499                 old_state == NVME_CTRL_RESETTING) {
 500                 nvme_start_failfast_work(ctrl);
 501         }
 502         return changed;
 503 }
 504 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 505
 506 /*
 507  * Returns true for sink states that can't ever transition back to live.
 508  */
 509 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 510 {
 511         switch (ctrl->state) {
 512         case NVME_CTRL_NEW:
 513         case NVME_CTRL_LIVE:
 514         case NVME_CTRL_RESETTING:
 515         case NVME_CTRL_CONNECTING:
 516                 return false;
 517         case NVME_CTRL_DELETING:
 518         case NVME_CTRL_DELETING_NOIO:
 519         case NVME_CTRL_DEAD:
 520                 return true;
 521         default:
 522                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 523                 return true;
 524         }
 525 }
 526
 527 /*
 528  * Waits for the controller state to be resetting, or returns false if it is
 529  * not possible to ever transition to that state.
 530  */
 531 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 532 {
 533         wait_event(ctrl->state_wq,
 534                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 535                    nvme_state_terminal(ctrl));
 536         return ctrl->state == NVME_CTRL_RESETTING;
 537 }
 538 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 539
 540 static void nvme_free_ns_head(struct kref *ref)
 541 {
 542         struct nvme_ns_head *head =
 543                 container_of(ref, struct nvme_ns_head, ref);
 544
 545         nvme_mpath_remove_disk(head);
 546         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 547         cleanup_srcu_struct(&head->srcu);
 548         nvme_put_subsystem(head->subsys);
 549         kfree(head);
 550 }
 551
 552 static void nvme_put_ns_head(struct nvme_ns_head *head)
 553 {
 554         kref_put(&head->ref, nvme_free_ns_head);
 555 }
 556
 557 static void nvme_free_ns(struct kref *kref)
 558 {
 559         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 560
 561         if (ns->ndev)
 562                 nvme_nvm_unregister(ns);
 563
 564         put_disk(ns->disk);
 565         nvme_put_ns_head(ns->head);
 566         nvme_put_ctrl(ns->ctrl);
 567         kfree(ns);
 568 }
 569
 570 void nvme_put_ns(struct nvme_ns *ns)
 571 {
 572         kref_put(&ns->kref, nvme_free_ns);
 573 }
 574 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 575
 576 static inline void nvme_clear_nvme_request(struct request *req)
 577 {
 578         if (!(req->rq_flags & RQF_DONTPREP)) {
 579                 nvme_req(req)->retries = 0;
 580                 nvme_req(req)->flags = 0;
 581                 req->rq_flags |= RQF_DONTPREP;
 582         }
 583 }
 584
 585 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 586 {
 587         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 588 }
 589
 590 static inline void nvme_init_request(struct request *req,
 591                 struct nvme_command *cmd)
 592 {
 593         if (req->q->queuedata)
 594                 req->timeout = NVME_IO_TIMEOUT;
 595         else /* no queuedata implies admin queue */
 596                 req->timeout = NVME_ADMIN_TIMEOUT;
 597
 598         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 599         nvme_clear_nvme_request(req);
 600         nvme_req(req)->cmd = cmd;
 601 }
 602
 603 struct request *nvme_alloc_request(struct request_queue *q,
 604                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 605 {
 606         struct request *req;
 607
 608         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 609         if (!IS_ERR(req))
 610                 nvme_init_request(req, cmd);
 611         return req;
 612 }
 613 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 614
 615 static struct request *nvme_alloc_request_qid(struct request_queue *q,
 616                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 617 {
 618         struct request *req;
 619
 620         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 621                         qid ? qid - 1 : 0);
 622         if (!IS_ERR(req))
 623                 nvme_init_request(req, cmd);
 624         return req;
 625 }
 626
 627 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 628 {
 629         struct nvme_command c;
 630
 631         memset(&c, 0, sizeof(c));
 632
 633         c.directive.opcode = nvme_admin_directive_send;
 634         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 635         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 636         c.directive.dtype = NVME_DIR_IDENTIFY;
 637         c.directive.tdtype = NVME_DIR_STREAMS;
 638         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 639
 640         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 641 }
 642
 643 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 644 {
 645         return nvme_toggle_streams(ctrl, false);
 646 }
 647
 648 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 649 {
 650         return nvme_toggle_streams(ctrl, true);
 651 }
 652
 653 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 654                                   struct streams_directive_params *s, u32 nsid)
 655 {
 656         struct nvme_command c;
 657
 658         memset(&c, 0, sizeof(c));
 659         memset(s, 0, sizeof(*s));
 660
 661         c.directive.opcode = nvme_admin_directive_recv;
 662         c.directive.nsid = cpu_to_le32(nsid);
 663         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 664         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 665         c.directive.dtype = NVME_DIR_STREAMS;
 666
 667         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 668 }
 669
 670 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 671 {
 672         struct streams_directive_params s;
 673         int ret;
 674
 675         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 676                 return 0;
 677         if (!streams)
 678                 return 0;
 679
 680         ret = nvme_enable_streams(ctrl);
 681         if (ret)
 682                 return ret;
 683
 684         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 685         if (ret)
 686                 goto out_disable_stream;
 687
 688         ctrl->nssa = le16_to_cpu(s.nssa);
 689         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 690                 dev_info(ctrl->device, "too few streams (%u) available\n",
 691                                         ctrl->nssa);
 692                 goto out_disable_stream;
 693         }
 694
 695         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 696         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 697         return 0;
 698
 699 out_disable_stream:
 700         nvme_disable_streams(ctrl);
 701         return ret;
 702 }
 703
 704 /*
 705  * Check if 'req' has a write hint associated with it. If it does, assign
 706  * a valid namespace stream to the write.
 707  */
 708 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 709                                      struct request *req, u16 *control,
 710                                      u32 *dsmgmt)
 711 {
 712         enum rw_hint streamid = req->write_hint;
 713
 714         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 715                 streamid = 0;
 716         else {
 717                 streamid--;
 718                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 719                         return;
 720
 721                 *control |= NVME_RW_DTYPE_STREAMS;
 722                 *dsmgmt |= streamid << 16;
 723         }
 724
 725         if (streamid < ARRAY_SIZE(req->q->write_hints))
 726                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 727 }
 728
 729 static void nvme_setup_passthrough(struct request *req,
 730                 struct nvme_command *cmd)
 731 {
 732         memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 733         /* passthru commands should let the driver set the SGL flags */
 734         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 735 }
 736
 737 static inline void nvme_setup_flush(struct nvme_ns *ns,
 738                 struct nvme_command *cmnd)
 739 {
 740         cmnd->common.opcode = nvme_cmd_flush;
 741         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 742 }
 743
 744 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 745                 struct nvme_command *cmnd)
 746 {
 747         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 748         struct nvme_dsm_range *range;
 749         struct bio *bio;
 750
 751         /*
 752          * Some devices do not consider the DSM 'Number of Ranges' field when
 753          * determining how much data to DMA. Always allocate memory for maximum
 754          * number of segments to prevent device reading beyond end of buffer.
 755          */
 756         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 757
 758         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 759         if (!range) {
 760                 /*
 761                  * If we fail allocation our range, fallback to the controller
 762                  * discard page. If that's also busy, it's safe to return
 763                  * busy, as we know we can make progress once that's freed.
 764                  */
 765                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 766                         return BLK_STS_RESOURCE;
 767
 768                 range = page_address(ns->ctrl->discard_page);
 769         }
 770
 771         __rq_for_each_bio(bio, req) {
 772                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 773                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 774
 775                 if (n < segments) {
 776                         range[n].cattr = cpu_to_le32(0);
 777                         range[n].nlb = cpu_to_le32(nlb);
 778                         range[n].slba = cpu_to_le64(slba);
 779                 }
 780                 n++;
 781         }
 782
 783         if (WARN_ON_ONCE(n != segments)) {
 784                 if (virt_to_page(range) == ns->ctrl->discard_page)
 785                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 786                 else
 787                         kfree(range);
 788                 return BLK_STS_IOERR;
 789         }
 790
 791         cmnd->dsm.opcode = nvme_cmd_dsm;
 792         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 793         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 794         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 795
 796         req->special_vec.bv_page = virt_to_page(range);
 797         req->special_vec.bv_offset = offset_in_page(range);
 798         req->special_vec.bv_len = alloc_size;
 799         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 800
 801         return BLK_STS_OK;
 802 }
 803
 804 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 805                 struct request *req, struct nvme_command *cmnd)
 806 {
 807         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 808                 return nvme_setup_discard(ns, req, cmnd);
 809
 810         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 811         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 812         cmnd->write_zeroes.slba =
 813                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 814         cmnd->write_zeroes.length =
 815                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 816         cmnd->write_zeroes.control = 0;
 817         return BLK_STS_OK;
 818 }
 819
 820 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 821                 struct request *req, struct nvme_command *cmnd,
 822                 enum nvme_opcode op)
 823 {
 824         struct nvme_ctrl *ctrl = ns->ctrl;
 825         u16 control = 0;
 826         u32 dsmgmt = 0;
 827
 828         if (req->cmd_flags & REQ_FUA)
 829                 control |= NVME_RW_FUA;
 830         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 831                 control |= NVME_RW_LR;
 832
 833         if (req->cmd_flags & REQ_RAHEAD)
 834                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 835
 836         cmnd->rw.opcode = op;
 837         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 838         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 839         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 840
 841         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 842                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 843
 844         if (ns->ms) {
 845                 /*
 846                  * If formated with metadata, the block layer always provides a
 847                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 848                  * we enable the PRACT bit for protection information or set the
 849                  * namespace capacity to zero to prevent any I/O.
 850                  */
 851                 if (!blk_integrity_rq(req)) {
 852                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 853                                 return BLK_STS_NOTSUPP;
 854                         control |= NVME_RW_PRINFO_PRACT;
 855                 }
 856
 857                 switch (ns->pi_type) {
 858                 case NVME_NS_DPS_PI_TYPE3:
 859                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 860                         break;
 861                 case NVME_NS_DPS_PI_TYPE1:
 862                 case NVME_NS_DPS_PI_TYPE2:
 863                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 864                                         NVME_RW_PRINFO_PRCHK_REF;
 865                         if (op == nvme_cmd_zone_append)
 866                                 control |= NVME_RW_APPEND_PIREMAP;
 867                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 868                         break;
 869                 }
 870         }
 871
 872         cmnd->rw.control = cpu_to_le16(control);
 873         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 874         return 0;
 875 }
 876
 877 void nvme_cleanup_cmd(struct request *req)
 878 {
 879         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 880                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 881                 struct page *page = req->special_vec.bv_page;
 882
 883                 if (page == ctrl->discard_page)
 884                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 885                 else
 886                         kfree(page_address(page) + req->special_vec.bv_offset);
 887         }
 888 }
 889 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 890
 891 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 892                 struct nvme_command *cmd)
 893 {
 894         blk_status_t ret = BLK_STS_OK;
 895
 896         nvme_clear_nvme_request(req);
 897
 898         memset(cmd, 0, sizeof(*cmd));
 899         switch (req_op(req)) {
 900         case REQ_OP_DRV_IN:
 901         case REQ_OP_DRV_OUT:
 902                 nvme_setup_passthrough(req, cmd);
 903                 break;
 904         case REQ_OP_FLUSH:
 905                 nvme_setup_flush(ns, cmd);
 906                 break;
 907         case REQ_OP_ZONE_RESET_ALL:
 908         case REQ_OP_ZONE_RESET:
 909                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 910                 break;
 911         case REQ_OP_ZONE_OPEN:
 912                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 913                 break;
 914         case REQ_OP_ZONE_CLOSE:
 915                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 916                 break;
 917         case REQ_OP_ZONE_FINISH:
 918                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 919                 break;
 920         case REQ_OP_WRITE_ZEROES:
 921                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 922                 break;
 923         case REQ_OP_DISCARD:
 924                 ret = nvme_setup_discard(ns, req, cmd);
 925                 break;
 926         case REQ_OP_READ:
 927                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 928                 break;
 929         case REQ_OP_WRITE:
 930                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 931                 break;
 932         case REQ_OP_ZONE_APPEND:
 933                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 934                 break;
 935         default:
 936                 WARN_ON_ONCE(1);
 937                 return BLK_STS_IOERR;
 938         }
 939
 940         cmd->common.command_id = req->tag;
 941         trace_nvme_setup_cmd(req, cmd);
 942         return ret;
 943 }
 944 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 945
 946 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 947 {
 948         struct completion *waiting = rq->end_io_data;
 949
 950         rq->end_io_data = NULL;
 951         complete(waiting);
 952 }
 953
 954 static void nvme_execute_rq_polled(struct request_queue *q,
 955                 struct gendisk *bd_disk, struct request *rq, int at_head)
 956 {
 957         DECLARE_COMPLETION_ONSTACK(wait);
 958
 959         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 960
 961         rq->cmd_flags |= REQ_HIPRI;
 962         rq->end_io_data = &wait;
 963         blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
 964
 965         while (!completion_done(&wait)) {
 966                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 967                 cond_resched();
 968         }
 969 }
 970
 971 /*
 972  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 973  * if the result is positive, it's an NVM Express status code
 974  */
 975 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 976                 union nvme_result *result, void *buffer, unsigned bufflen,
 977                 unsigned timeout, int qid, int at_head,
 978                 blk_mq_req_flags_t flags, bool poll)
 979 {
 980         struct request *req;
 981         int ret;
 982
 983         if (qid == NVME_QID_ANY)
 984                 req = nvme_alloc_request(q, cmd, flags);
 985         else
 986                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
 987         if (IS_ERR(req))
 988                 return PTR_ERR(req);
 989
 990         if (timeout)
 991                 req->timeout = timeout;
 992
 993         if (buffer && bufflen) {
 994                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 995                 if (ret)
 996                         goto out;
 997         }
 998
 999         if (poll)
1000                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
1001         else
1002                 blk_execute_rq(NULL, req, at_head);
1003         if (result)
1004                 *result = nvme_req(req)->result;
1005         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1006                 ret = -EINTR;
1007         else
1008                 ret = nvme_req(req)->status;
1009  out:
1010         blk_mq_free_request(req);
1011         return ret;
1012 }
1013 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1014
1015 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1016                 void *buffer, unsigned bufflen)
1017 {
1018         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1019                         NVME_QID_ANY, 0, 0, false);
1020 }
1021 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1022
1023 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
1024                 unsigned len, u32 seed, bool write)
1025 {
1026         struct bio_integrity_payload *bip;
1027         int ret = -ENOMEM;
1028         void *buf;
1029
1030         buf = kmalloc(len, GFP_KERNEL);
1031         if (!buf)
1032                 goto out;
1033
1034         ret = -EFAULT;
1035         if (write && copy_from_user(buf, ubuf, len))
1036                 goto out_free_meta;
1037
1038         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1039         if (IS_ERR(bip)) {
1040                 ret = PTR_ERR(bip);
1041                 goto out_free_meta;
1042         }
1043
1044         bip->bip_iter.bi_size = len;
1045         bip->bip_iter.bi_sector = seed;
1046         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1047                         offset_in_page(buf));
1048         if (ret == len)
1049                 return buf;
1050         ret = -ENOMEM;
1051 out_free_meta:
1052         kfree(buf);
1053 out:
1054         return ERR_PTR(ret);
1055 }
1056
1057 static u32 nvme_known_admin_effects(u8 opcode)
1058 {
1059         switch (opcode) {
1060         case nvme_admin_format_nvm:
1061                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1062                         NVME_CMD_EFFECTS_CSE_MASK;
1063         case nvme_admin_sanitize_nvm:
1064                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1065         default:
1066                 break;
1067         }
1068         return 0;
1069 }
1070
1071 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1072 {
1073         u32 effects = 0;
1074
1075         if (ns) {
1076                 if (ns->head->effects)
1077                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1078                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1079                         dev_warn(ctrl->device,
1080                                  "IO command:%02x has unhandled effects:%08x\n",
1081                                  opcode, effects);
1082                 return 0;
1083         }
1084
1085         if (ctrl->effects)
1086                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1087         effects |= nvme_known_admin_effects(opcode);
1088
1089         return effects;
1090 }
1091 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1092
1093 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1094                                u8 opcode)
1095 {
1096         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1097
1098         /*
1099          * For simplicity, IO to all namespaces is quiesced even if the command
1100          * effects say only one namespace is affected.
1101          */
1102         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1103                 mutex_lock(&ctrl->scan_lock);
1104                 mutex_lock(&ctrl->subsys->lock);
1105                 nvme_mpath_start_freeze(ctrl->subsys);
1106                 nvme_mpath_wait_freeze(ctrl->subsys);
1107                 nvme_start_freeze(ctrl);
1108                 nvme_wait_freeze(ctrl);
1109         }
1110         return effects;
1111 }
1112
1113 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1114 {
1115         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1116                 nvme_unfreeze(ctrl);
1117                 nvme_mpath_unfreeze(ctrl->subsys);
1118                 mutex_unlock(&ctrl->subsys->lock);
1119                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1120                 mutex_unlock(&ctrl->scan_lock);
1121         }
1122         if (effects & NVME_CMD_EFFECTS_CCC)
1123                 nvme_init_identify(ctrl);
1124         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1125                 nvme_queue_scan(ctrl);
1126                 flush_work(&ctrl->scan_work);
1127         }
1128 }
1129
1130 void nvme_execute_passthru_rq(struct request *rq)
1131 {
1132         struct nvme_command *cmd = nvme_req(rq)->cmd;
1133         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1134         struct nvme_ns *ns = rq->q->queuedata;
1135         struct gendisk *disk = ns ? ns->disk : NULL;
1136         u32 effects;
1137
1138         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1139         blk_execute_rq(disk, rq, 0);
1140         nvme_passthru_end(ctrl, effects);
1141 }
1142 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1143
1144 static int nvme_submit_user_cmd(struct request_queue *q,
1145                 struct nvme_command *cmd, void __user *ubuffer,
1146                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1147                 u32 meta_seed, u64 *result, unsigned timeout)
1148 {
1149         bool write = nvme_is_write(cmd);
1150         struct nvme_ns *ns = q->queuedata;
1151         struct block_device *bdev = ns ? ns->disk->part0 : NULL;
1152         struct request *req;
1153         struct bio *bio = NULL;
1154         void *meta = NULL;
1155         int ret;
1156
1157         req = nvme_alloc_request(q, cmd, 0);
1158         if (IS_ERR(req))
1159                 return PTR_ERR(req);
1160
1161         if (timeout)
1162                 req->timeout = timeout;
1163         nvme_req(req)->flags |= NVME_REQ_USERCMD;
1164
1165         if (ubuffer && bufflen) {
1166                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1167                                 GFP_KERNEL);
1168                 if (ret)
1169                         goto out;
1170                 bio = req->bio;
1171                 if (bdev)
1172                         bio_set_dev(bio, bdev);
1173                 if (bdev && meta_buffer && meta_len) {
1174                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1175                                         meta_seed, write);
1176                         if (IS_ERR(meta)) {
1177                                 ret = PTR_ERR(meta);
1178                                 goto out_unmap;
1179                         }
1180                         req->cmd_flags |= REQ_INTEGRITY;
1181                 }
1182         }
1183
1184         nvme_execute_passthru_rq(req);
1185         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1186                 ret = -EINTR;
1187         else
1188                 ret = nvme_req(req)->status;
1189         if (result)
1190                 *result = le64_to_cpu(nvme_req(req)->result.u64);
1191         if (meta && !ret && !write) {
1192                 if (copy_to_user(meta_buffer, meta, meta_len))
1193                         ret = -EFAULT;
1194         }
1195         kfree(meta);
1196  out_unmap:
1197         if (bio)
1198                 blk_rq_unmap_user(bio);
1199  out:
1200         blk_mq_free_request(req);
1201         return ret;
1202 }
1203
1204 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1205 {
1206         struct nvme_ctrl *ctrl = rq->end_io_data;
1207         unsigned long flags;
1208         bool startka = false;
1209
1210         blk_mq_free_request(rq);
1211
1212         if (status) {
1213                 dev_err(ctrl->device,
1214                         "failed nvme_keep_alive_end_io error=%d\n",
1215                                 status);
1216                 return;
1217         }
1218
1219         ctrl->comp_seen = false;
1220         spin_lock_irqsave(&ctrl->lock, flags);
1221         if (ctrl->state == NVME_CTRL_LIVE ||
1222             ctrl->state == NVME_CTRL_CONNECTING)
1223                 startka = true;
1224         spin_unlock_irqrestore(&ctrl->lock, flags);
1225         if (startka)
1226                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1227 }
1228
1229 static void nvme_keep_alive_work(struct work_struct *work)
1230 {
1231         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1232                         struct nvme_ctrl, ka_work);
1233         bool comp_seen = ctrl->comp_seen;
1234         struct request *rq;
1235
1236         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1237                 dev_dbg(ctrl->device,
1238                         "reschedule traffic based keep-alive timer\n");
1239                 ctrl->comp_seen = false;
1240                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1241                 return;
1242         }
1243
1244         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1245                                 BLK_MQ_REQ_RESERVED);
1246         if (IS_ERR(rq)) {
1247                 /* allocation failure, reset the controller */
1248                 dev_err(ctrl->device, "keep-alive failed\n");
1249                 nvme_reset_ctrl(ctrl);
1250                 return;
1251         }
1252
1253         rq->timeout = ctrl->kato * HZ;
1254         rq->end_io_data = ctrl;
1255         blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1256 }
1257
1258 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1259 {
1260         if (unlikely(ctrl->kato == 0))
1261                 return;
1262
1263         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1264 }
1265
1266 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1267 {
1268         if (unlikely(ctrl->kato == 0))
1269                 return;
1270
1271         cancel_delayed_work_sync(&ctrl->ka_work);
1272 }
1273 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1274
1275 /*
1276  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1277  * flag, thus sending any new CNS opcodes has a big chance of not working.
1278  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1279  * (but not for any later version).
1280  */
1281 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1282 {
1283         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1284                 return ctrl->vs < NVME_VS(1, 2, 0);
1285         return ctrl->vs < NVME_VS(1, 1, 0);
1286 }
1287
1288 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1289 {
1290         struct nvme_command c = { };
1291         int error;
1292
1293         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1294         c.identify.opcode = nvme_admin_identify;
1295         c.identify.cns = NVME_ID_CNS_CTRL;
1296
1297         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1298         if (!*id)
1299                 return -ENOMEM;
1300
1301         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1302                         sizeof(struct nvme_id_ctrl));
1303         if (error)
1304                 kfree(*id);
1305         return error;
1306 }
1307
1308 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1309 {
1310         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1311 }
1312
1313 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1314                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1315 {
1316         const char *warn_str = "ctrl returned bogus length:";
1317         void *data = cur;
1318
1319         switch (cur->nidt) {
1320         case NVME_NIDT_EUI64:
1321                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1322                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1323                                  warn_str, cur->nidl);
1324                         return -1;
1325                 }
1326                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1327                 return NVME_NIDT_EUI64_LEN;
1328         case NVME_NIDT_NGUID:
1329                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1330                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1331                                  warn_str, cur->nidl);
1332                         return -1;
1333                 }
1334                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1335                 return NVME_NIDT_NGUID_LEN;
1336         case NVME_NIDT_UUID:
1337                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1338                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1339                                  warn_str, cur->nidl);
1340                         return -1;
1341                 }
1342                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1343                 return NVME_NIDT_UUID_LEN;
1344         case NVME_NIDT_CSI:
1345                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1346                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1347                                  warn_str, cur->nidl);
1348                         return -1;
1349                 }
1350                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1351                 *csi_seen = true;
1352                 return NVME_NIDT_CSI_LEN;
1353         default:
1354                 /* Skip unknown types */
1355                 return cur->nidl;
1356         }
1357 }
1358
1359 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1360                 struct nvme_ns_ids *ids)
1361 {
1362         struct nvme_command c = { };
1363         bool csi_seen = false;
1364         int status, pos, len;
1365         void *data;
1366
1367         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1368                 return 0;
1369         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1370                 return 0;
1371
1372         c.identify.opcode = nvme_admin_identify;
1373         c.identify.nsid = cpu_to_le32(nsid);
1374         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1375
1376         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1377         if (!data)
1378                 return -ENOMEM;
1379
1380         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1381                                       NVME_IDENTIFY_DATA_SIZE);
1382         if (status) {
1383                 dev_warn(ctrl->device,
1384                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1385                         nsid, status);
1386                 goto free_data;
1387         }
1388
1389         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1390                 struct nvme_ns_id_desc *cur = data + pos;
1391
1392                 if (cur->nidl == 0)
1393                         break;
1394
1395                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1396                 if (len < 0)
1397                         break;
1398
1399                 len += sizeof(*cur);
1400         }
1401
1402         if (nvme_multi_css(ctrl) && !csi_seen) {
1403                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1404                          nsid);
1405                 status = -EINVAL;
1406         }
1407
1408 free_data:
1409         kfree(data);
1410         return status;
1411 }
1412
1413 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1414                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1415 {
1416         struct nvme_command c = { };
1417         int error;
1418
1419         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1420         c.identify.opcode = nvme_admin_identify;
1421         c.identify.nsid = cpu_to_le32(nsid);
1422         c.identify.cns = NVME_ID_CNS_NS;
1423
1424         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1425         if (!*id)
1426                 return -ENOMEM;
1427
1428         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1429         if (error) {
1430                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1431                 goto out_free_id;
1432         }
1433
1434         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1435         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1436                 goto out_free_id;
1437
1438         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1439             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1440                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1441         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1442             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1443                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1444
1445         return 0;
1446
1447 out_free_id:
1448         kfree(*id);
1449         return error;
1450 }
1451
1452 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1453                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1454 {
1455         union nvme_result res = { 0 };
1456         struct nvme_command c;
1457         int ret;
1458
1459         memset(&c, 0, sizeof(c));
1460         c.features.opcode = op;
1461         c.features.fid = cpu_to_le32(fid);
1462         c.features.dword11 = cpu_to_le32(dword11);
1463
1464         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1465                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1466         if (ret >= 0 && result)
1467                 *result = le32_to_cpu(res.u32);
1468         return ret;
1469 }
1470
1471 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1472                       unsigned int dword11, void *buffer, size_t buflen,
1473                       u32 *result)
1474 {
1475         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1476                              buflen, result);
1477 }
1478 EXPORT_SYMBOL_GPL(nvme_set_features);
1479
1480 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1481                       unsigned int dword11, void *buffer, size_t buflen,
1482                       u32 *result)
1483 {
1484         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1485                              buflen, result);
1486 }
1487 EXPORT_SYMBOL_GPL(nvme_get_features);
1488
1489 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1490 {
1491         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1492         u32 result;
1493         int status, nr_io_queues;
1494
1495         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1496                         &result);
1497         if (status < 0)
1498                 return status;
1499
1500         /*
1501          * Degraded controllers might return an error when setting the queue
1502          * count.  We still want to be able to bring them online and offer
1503          * access to the admin queue, as that might be only way to fix them up.
1504          */
1505         if (status > 0) {
1506                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1507                 *count = 0;
1508         } else {
1509                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1510                 *count = min(*count, nr_io_queues);
1511         }
1512
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1516
1517 #define NVME_AEN_SUPPORTED \
1518         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1519          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1520
1521 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1522 {
1523         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1524         int status;
1525
1526         if (!supported_aens)
1527                 return;
1528
1529         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1530                         NULL, 0, &result);
1531         if (status)
1532                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1533                          supported_aens);
1534
1535         queue_work(nvme_wq, &ctrl->async_event_work);
1536 }
1537
1538 /*
1539  * Convert integer values from ioctl structures to user pointers, silently
1540  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1541  * kernels.
1542  */
1543 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1544 {
1545         if (in_compat_syscall())
1546                 ptrval = (compat_uptr_t)ptrval;
1547         return (void __user *)ptrval;
1548 }
1549
1550 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1551 {
1552         struct nvme_user_io io;
1553         struct nvme_command c;
1554         unsigned length, meta_len;
1555         void __user *metadata;
1556
1557         if (copy_from_user(&io, uio, sizeof(io)))
1558                 return -EFAULT;
1559         if (io.flags)
1560                 return -EINVAL;
1561
1562         switch (io.opcode) {
1563         case nvme_cmd_write:
1564         case nvme_cmd_read:
1565         case nvme_cmd_compare:
1566                 break;
1567         default:
1568                 return -EINVAL;
1569         }
1570
1571         length = (io.nblocks + 1) << ns->lba_shift;
1572
1573         if ((io.control & NVME_RW_PRINFO_PRACT) &&
1574             ns->ms == sizeof(struct t10_pi_tuple)) {
1575                 /*
1576                  * Protection information is stripped/inserted by the
1577                  * controller.
1578                  */
1579                 if (nvme_to_user_ptr(io.metadata))
1580                         return -EINVAL;
1581                 meta_len = 0;
1582                 metadata = NULL;
1583         } else {
1584                 meta_len = (io.nblocks + 1) * ns->ms;
1585                 metadata = nvme_to_user_ptr(io.metadata);
1586         }
1587
1588         if (ns->features & NVME_NS_EXT_LBAS) {
1589                 length += meta_len;
1590                 meta_len = 0;
1591         } else if (meta_len) {
1592                 if ((io.metadata & 3) || !io.metadata)
1593                         return -EINVAL;
1594         }
1595
1596         memset(&c, 0, sizeof(c));
1597         c.rw.opcode = io.opcode;
1598         c.rw.flags = io.flags;
1599         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1600         c.rw.slba = cpu_to_le64(io.slba);
1601         c.rw.length = cpu_to_le16(io.nblocks);
1602         c.rw.control = cpu_to_le16(io.control);
1603         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1604         c.rw.reftag = cpu_to_le32(io.reftag);
1605         c.rw.apptag = cpu_to_le16(io.apptag);
1606         c.rw.appmask = cpu_to_le16(io.appmask);
1607
1608         return nvme_submit_user_cmd(ns->queue, &c,
1609                         nvme_to_user_ptr(io.addr), length,
1610                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1611 }
1612
1613 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1614                         struct nvme_passthru_cmd __user *ucmd)
1615 {
1616         struct nvme_passthru_cmd cmd;
1617         struct nvme_command c;
1618         unsigned timeout = 0;
1619         u64 result;
1620         int status;
1621
1622         if (!capable(CAP_SYS_ADMIN))
1623                 return -EACCES;
1624         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1625                 return -EFAULT;
1626         if (cmd.flags)
1627                 return -EINVAL;
1628
1629         memset(&c, 0, sizeof(c));
1630         c.common.opcode = cmd.opcode;
1631         c.common.flags = cmd.flags;
1632         c.common.nsid = cpu_to_le32(cmd.nsid);
1633         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1634         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1635         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1636         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1637         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1638         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1639         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1640         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1641
1642         if (cmd.timeout_ms)
1643                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1644
1645         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1646                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1647                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1648                         0, &result, timeout);
1649
1650         if (status >= 0) {
1651                 if (put_user(result, &ucmd->result))
1652                         return -EFAULT;
1653         }
1654
1655         return status;
1656 }
1657
1658 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1659                         struct nvme_passthru_cmd64 __user *ucmd)
1660 {
1661         struct nvme_passthru_cmd64 cmd;
1662         struct nvme_command c;
1663         unsigned timeout = 0;
1664         int status;
1665
1666         if (!capable(CAP_SYS_ADMIN))
1667                 return -EACCES;
1668         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1669                 return -EFAULT;
1670         if (cmd.flags)
1671                 return -EINVAL;
1672
1673         memset(&c, 0, sizeof(c));
1674         c.common.opcode = cmd.opcode;
1675         c.common.flags = cmd.flags;
1676         c.common.nsid = cpu_to_le32(cmd.nsid);
1677         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1678         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1679         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1680         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1681         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1682         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1683         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1684         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1685
1686         if (cmd.timeout_ms)
1687                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1688
1689         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1690                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1691                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1692                         0, &cmd.result, timeout);
1693
1694         if (status >= 0) {
1695                 if (put_user(cmd.result, &ucmd->result))
1696                         return -EFAULT;
1697         }
1698
1699         return status;
1700 }
1701
1702 /*
1703  * Issue ioctl requests on the first available path.  Note that unlike normal
1704  * block layer requests we will not retry failed request on another controller.
1705  */
1706 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1707                 struct nvme_ns_head **head, int *srcu_idx)
1708 {
1709 #ifdef CONFIG_NVME_MULTIPATH
1710         if (disk->fops == &nvme_ns_head_ops) {
1711                 struct nvme_ns *ns;
1712
1713                 *head = disk->private_data;
1714                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1715                 ns = nvme_find_path(*head);
1716                 if (!ns)
1717                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1718                 return ns;
1719         }
1720 #endif
1721         *head = NULL;
1722         *srcu_idx = -1;
1723         return disk->private_data;
1724 }
1725
1726 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1727 {
1728         if (head)
1729                 srcu_read_unlock(&head->srcu, idx);
1730 }
1731
1732 static bool is_ctrl_ioctl(unsigned int cmd)
1733 {
1734         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1735                 return true;
1736         if (is_sed_ioctl(cmd))
1737                 return true;
1738         return false;
1739 }
1740
1741 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1742                                   void __user *argp,
1743                                   struct nvme_ns_head *head,
1744                                   int srcu_idx)
1745 {
1746         struct nvme_ctrl *ctrl = ns->ctrl;
1747         int ret;
1748
1749         nvme_get_ctrl(ns->ctrl);
1750         nvme_put_ns_from_disk(head, srcu_idx);
1751
1752         switch (cmd) {
1753         case NVME_IOCTL_ADMIN_CMD:
1754                 ret = nvme_user_cmd(ctrl, NULL, argp);
1755                 break;
1756         case NVME_IOCTL_ADMIN64_CMD:
1757                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1758                 break;
1759         default:
1760                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1761                 break;
1762         }
1763         nvme_put_ctrl(ctrl);
1764         return ret;
1765 }
1766
1767 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1768                 unsigned int cmd, unsigned long arg)
1769 {
1770         struct nvme_ns_head *head = NULL;
1771         void __user *argp = (void __user *)arg;
1772         struct nvme_ns *ns;
1773         int srcu_idx, ret;
1774
1775         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1776         if (unlikely(!ns))
1777                 return -EWOULDBLOCK;
1778
1779         /*
1780          * Handle ioctls that apply to the controller instead of the namespace
1781          * seperately and drop the ns SRCU reference early.  This avoids a
1782          * deadlock when deleting namespaces using the passthrough interface.
1783          */
1784         if (is_ctrl_ioctl(cmd))
1785                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1786
1787         switch (cmd) {
1788         case NVME_IOCTL_ID:
1789                 force_successful_syscall_return();
1790                 ret = ns->head->ns_id;
1791                 break;
1792         case NVME_IOCTL_IO_CMD:
1793                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1794                 break;
1795         case NVME_IOCTL_SUBMIT_IO:
1796                 ret = nvme_submit_io(ns, argp);
1797                 break;
1798         case NVME_IOCTL_IO64_CMD:
1799                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1800                 break;
1801         default:
1802                 if (ns->ndev)
1803                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1804                 else
1805                         ret = -ENOTTY;
1806         }
1807
1808         nvme_put_ns_from_disk(head, srcu_idx);
1809         return ret;
1810 }
1811
1812 #ifdef CONFIG_COMPAT
1813 struct nvme_user_io32 {
1814         __u8    opcode;
1815         __u8    flags;
1816         __u16   control;
1817         __u16   nblocks;
1818         __u16   rsvd;
1819         __u64   metadata;
1820         __u64   addr;
1821         __u64   slba;
1822         __u32   dsmgmt;
1823         __u32   reftag;
1824         __u16   apptag;
1825         __u16   appmask;
1826 } __attribute__((__packed__));
1827
1828 #define NVME_IOCTL_SUBMIT_IO32  _IOW('N', 0x42, struct nvme_user_io32)
1829
1830 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1831                 unsigned int cmd, unsigned long arg)
1832 {
1833         /*
1834          * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1835          * between 32 bit programs and 64 bit kernel.
1836          * The cause is that the results of sizeof(struct nvme_user_io),
1837          * which is used to define NVME_IOCTL_SUBMIT_IO,
1838          * are not same between 32 bit compiler and 64 bit compiler.
1839          * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1840          * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1841          * Other IOCTL numbers are same between 32 bit and 64 bit.
1842          * So there is nothing to do regarding to other IOCTL numbers.
1843          */
1844         if (cmd == NVME_IOCTL_SUBMIT_IO32)
1845                 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1846
1847         return nvme_ioctl(bdev, mode, cmd, arg);
1848 }
1849 #else
1850 #define nvme_compat_ioctl       NULL
1851 #endif /* CONFIG_COMPAT */
1852
1853 static int nvme_open(struct block_device *bdev, fmode_t mode)
1854 {
1855         struct nvme_ns *ns = bdev->bd_disk->private_data;
1856
1857 #ifdef CONFIG_NVME_MULTIPATH
1858         /* should never be called due to GENHD_FL_HIDDEN */
1859         if (WARN_ON_ONCE(ns->head->disk))
1860                 goto fail;
1861 #endif
1862         if (!kref_get_unless_zero(&ns->kref))
1863                 goto fail;
1864         if (!try_module_get(ns->ctrl->ops->module))
1865                 goto fail_put_ns;
1866
1867         return 0;
1868
1869 fail_put_ns:
1870         nvme_put_ns(ns);
1871 fail:
1872         return -ENXIO;
1873 }
1874
1875 static void nvme_release(struct gendisk *disk, fmode_t mode)
1876 {
1877         struct nvme_ns *ns = disk->private_data;
1878
1879         module_put(ns->ctrl->ops->module);
1880         nvme_put_ns(ns);
1881 }
1882
1883 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1884 {
1885         /* some standard values */
1886         geo->heads = 1 << 6;
1887         geo->sectors = 1 << 5;
1888         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1889         return 0;
1890 }
1891
1892 #ifdef CONFIG_BLK_DEV_INTEGRITY
1893 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1894                                 u32 max_integrity_segments)
1895 {
1896         struct blk_integrity integrity;
1897
1898         memset(&integrity, 0, sizeof(integrity));
1899         switch (pi_type) {
1900         case NVME_NS_DPS_PI_TYPE3:
1901                 integrity.profile = &t10_pi_type3_crc;
1902                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1903                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1904                 break;
1905         case NVME_NS_DPS_PI_TYPE1:
1906         case NVME_NS_DPS_PI_TYPE2:
1907                 integrity.profile = &t10_pi_type1_crc;
1908                 integrity.tag_size = sizeof(u16);
1909                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1910                 break;
1911         default:
1912                 integrity.profile = NULL;
1913                 break;
1914         }
1915         integrity.tuple_size = ms;
1916         blk_integrity_register(disk, &integrity);
1917         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1918 }
1919 #else
1920 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1921                                 u32 max_integrity_segments)
1922 {
1923 }
1924 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1925
1926 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1927 {
1928         struct nvme_ctrl *ctrl = ns->ctrl;
1929         struct request_queue *queue = disk->queue;
1930         u32 size = queue_logical_block_size(queue);
1931
1932         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1933                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1934                 return;
1935         }
1936
1937         if (ctrl->nr_streams && ns->sws && ns->sgs)
1938                 size *= ns->sws * ns->sgs;
1939
1940         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1941                         NVME_DSM_MAX_RANGES);
1942
1943         queue->limits.discard_alignment = 0;
1944         queue->limits.discard_granularity = size;
1945
1946         /* If discard is already enabled, don't reset queue limits */
1947         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1948                 return;
1949
1950         blk_queue_max_discard_sectors(queue, UINT_MAX);
1951         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1952
1953         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1954                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1955 }
1956
1957 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1958 {
1959         u64 max_blocks;
1960
1961         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1962             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1963                 return;
1964         /*
1965          * Even though NVMe spec explicitly states that MDTS is not
1966          * applicable to the write-zeroes:- "The restriction does not apply to
1967          * commands that do not transfer data between the host and the
1968          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1969          * In order to be more cautious use controller's max_hw_sectors value
1970          * to configure the maximum sectors for the write-zeroes which is
1971          * configured based on the controller's MDTS field in the
1972          * nvme_init_identify() if available.
1973          */
1974         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1975                 max_blocks = (u64)USHRT_MAX + 1;
1976         else
1977                 max_blocks = ns->ctrl->max_hw_sectors + 1;
1978
1979         blk_queue_max_write_zeroes_sectors(disk->queue,
1980                                            nvme_lba_to_sect(ns, max_blocks));
1981 }
1982
1983 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1984 {
1985         return !uuid_is_null(&ids->uuid) ||
1986                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1987                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1988 }
1989
1990 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1991 {
1992         return uuid_equal(&a->uuid, &b->uuid) &&
1993                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1994                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1995                 a->csi == b->csi;
1996 }
1997
1998 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1999                                  u32 *phys_bs, u32 *io_opt)
2000 {
2001         struct streams_directive_params s;
2002         int ret;
2003
2004         if (!ctrl->nr_streams)
2005                 return 0;
2006
2007         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2008         if (ret)
2009                 return ret;
2010
2011         ns->sws = le32_to_cpu(s.sws);
2012         ns->sgs = le16_to_cpu(s.sgs);
2013
2014         if (ns->sws) {
2015                 *phys_bs = ns->sws * (1 << ns->lba_shift);
2016                 if (ns->sgs)
2017                         *io_opt = *phys_bs * ns->sgs;
2018         }
2019
2020         return 0;
2021 }
2022
2023 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2024 {
2025         struct nvme_ctrl *ctrl = ns->ctrl;
2026
2027         /*
2028          * The PI implementation requires the metadata size to be equal to the
2029          * t10 pi tuple size.
2030          */
2031         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2032         if (ns->ms == sizeof(struct t10_pi_tuple))
2033                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2034         else
2035                 ns->pi_type = 0;
2036
2037         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2038         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2039                 return 0;
2040         if (ctrl->ops->flags & NVME_F_FABRICS) {
2041                 /*
2042                  * The NVMe over Fabrics specification only supports metadata as
2043                  * part of the extended data LBA.  We rely on HCA/HBA support to
2044                  * remap the separate metadata buffer from the block layer.
2045                  */
2046                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2047                         return -EINVAL;
2048                 if (ctrl->max_integrity_segments)
2049                         ns->features |=
2050                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2051         } else {
2052                 /*
2053                  * For PCIe controllers, we can't easily remap the separate
2054                  * metadata buffer from the block layer and thus require a
2055                  * separate metadata buffer for block layer metadata/PI support.
2056                  * We allow extended LBAs for the passthrough interface, though.
2057                  */
2058                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2059                         ns->features |= NVME_NS_EXT_LBAS;
2060                 else
2061                         ns->features |= NVME_NS_METADATA_SUPPORTED;
2062         }
2063
2064         return 0;
2065 }
2066
2067 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2068                 struct request_queue *q)
2069 {
2070         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2071
2072         if (ctrl->max_hw_sectors) {
2073                 u32 max_segments =
2074                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2075
2076                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2077                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2078                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2079         }
2080         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2081         blk_queue_dma_alignment(q, 7);
2082         blk_queue_write_cache(q, vwc, vwc);
2083 }
2084
2085 static void nvme_update_disk_info(struct gendisk *disk,
2086                 struct nvme_ns *ns, struct nvme_id_ns *id)
2087 {
2088         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2089         unsigned short bs = 1 << ns->lba_shift;
2090         u32 atomic_bs, phys_bs, io_opt = 0;
2091
2092         /*
2093          * The block layer can't support LBA sizes larger than the page size
2094          * yet, so catch this early and don't allow block I/O.
2095          */
2096         if (ns->lba_shift > PAGE_SHIFT) {
2097                 capacity = 0;
2098                 bs = (1 << 9);
2099         }
2100
2101         blk_integrity_unregister(disk);
2102
2103         atomic_bs = phys_bs = bs;
2104         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2105         if (id->nabo == 0) {
2106                 /*
2107                  * Bit 1 indicates whether NAWUPF is defined for this namespace
2108                  * and whether it should be used instead of AWUPF. If NAWUPF ==
2109                  * 0 then AWUPF must be used instead.
2110                  */
2111                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2112                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2113                 else
2114                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2115         }
2116
2117         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2118                 /* NPWG = Namespace Preferred Write Granularity */
2119                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2120                 /* NOWS = Namespace Optimal Write Size */
2121                 io_opt = bs * (1 + le16_to_cpu(id->nows));
2122         }
2123
2124         blk_queue_logical_block_size(disk->queue, bs);
2125         /*
2126          * Linux filesystems assume writing a single physical block is
2127          * an atomic operation. Hence limit the physical block size to the
2128          * value of the Atomic Write Unit Power Fail parameter.
2129          */
2130         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2131         blk_queue_io_min(disk->queue, phys_bs);
2132         blk_queue_io_opt(disk->queue, io_opt);
2133
2134         /*
2135          * Register a metadata profile for PI, or the plain non-integrity NVMe
2136          * metadata masquerading as Type 0 if supported, otherwise reject block
2137          * I/O to namespaces with metadata except when the namespace supports
2138          * PI, as it can strip/insert in that case.
2139          */
2140         if (ns->ms) {
2141                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2142                     (ns->features & NVME_NS_METADATA_SUPPORTED))
2143                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
2144                                             ns->ctrl->max_integrity_segments);
2145                 else if (!nvme_ns_has_pi(ns))
2146                         capacity = 0;
2147         }
2148
2149         set_capacity_and_notify(disk, capacity);
2150
2151         nvme_config_discard(disk, ns);
2152         nvme_config_write_zeroes(disk, ns);
2153
2154         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
2155                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
2156 }
2157
2158 static inline bool nvme_first_scan(struct gendisk *disk)
2159 {
2160         /* nvme_alloc_ns() scans the disk prior to adding it */
2161         return !(disk->flags & GENHD_FL_UP);
2162 }
2163
2164 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2165 {
2166         struct nvme_ctrl *ctrl = ns->ctrl;
2167         u32 iob;
2168
2169         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2170             is_power_of_2(ctrl->max_hw_sectors))
2171                 iob = ctrl->max_hw_sectors;
2172         else
2173                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2174
2175         if (!iob)
2176                 return;
2177
2178         if (!is_power_of_2(iob)) {
2179                 if (nvme_first_scan(ns->disk))
2180                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2181                                 ns->disk->disk_name, iob);
2182                 return;
2183         }
2184
2185         if (blk_queue_is_zoned(ns->disk->queue)) {
2186                 if (nvme_first_scan(ns->disk))
2187                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
2188                                 ns->disk->disk_name);
2189                 return;
2190         }
2191
2192         blk_queue_chunk_sectors(ns->queue, iob);
2193 }
2194
2195 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2196 {
2197         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2198         int ret;
2199
2200         blk_mq_freeze_queue(ns->disk->queue);
2201         ns->lba_shift = id->lbaf[lbaf].ds;
2202         nvme_set_queue_limits(ns->ctrl, ns->queue);
2203
2204         ret = nvme_configure_metadata(ns, id);
2205         if (ret)
2206                 goto out_unfreeze;
2207         nvme_set_chunk_sectors(ns, id);
2208         nvme_update_disk_info(ns->disk, ns, id);
2209
2210         if (ns->head->ids.csi == NVME_CSI_ZNS) {
2211                 ret = nvme_update_zone_info(ns, lbaf);
2212                 if (ret)
2213                         goto out_unfreeze;
2214         }
2215
2216         blk_mq_unfreeze_queue(ns->disk->queue);
2217
2218         if (blk_queue_is_zoned(ns->queue)) {
2219                 ret = nvme_revalidate_zones(ns);
2220                 if (ret && !nvme_first_scan(ns->disk))
2221                         return ret;
2222         }
2223
2224 #ifdef CONFIG_NVME_MULTIPATH
2225         if (ns->head->disk) {
2226                 blk_mq_freeze_queue(ns->head->disk->queue);
2227                 nvme_update_disk_info(ns->head->disk, ns, id);
2228                 blk_stack_limits(&ns->head->disk->queue->limits,
2229                                  &ns->queue->limits, 0);
2230                 blk_queue_update_readahead(ns->head->disk->queue);
2231                 blk_mq_unfreeze_queue(ns->head->disk->queue);
2232         }
2233 #endif
2234         return 0;
2235
2236 out_unfreeze:
2237         blk_mq_unfreeze_queue(ns->disk->queue);
2238         return ret;
2239 }
2240
2241 static char nvme_pr_type(enum pr_type type)
2242 {
2243         switch (type) {
2244         case PR_WRITE_EXCLUSIVE:
2245                 return 1;
2246         case PR_EXCLUSIVE_ACCESS:
2247                 return 2;
2248         case PR_WRITE_EXCLUSIVE_REG_ONLY:
2249                 return 3;
2250         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2251                 return 4;
2252         case PR_WRITE_EXCLUSIVE_ALL_REGS:
2253                 return 5;
2254         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2255                 return 6;
2256         default:
2257                 return 0;
2258         }
2259 };
2260
2261 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2262                                 u64 key, u64 sa_key, u8 op)
2263 {
2264         struct nvme_ns_head *head = NULL;
2265         struct nvme_ns *ns;
2266         struct nvme_command c;
2267         int srcu_idx, ret;
2268         u8 data[16] = { 0, };
2269
2270         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2271         if (unlikely(!ns))
2272                 return -EWOULDBLOCK;
2273
2274         put_unaligned_le64(key, &data[0]);
2275         put_unaligned_le64(sa_key, &data[8]);
2276
2277         memset(&c, 0, sizeof(c));
2278         c.common.opcode = op;
2279         c.common.nsid = cpu_to_le32(ns->head->ns_id);
2280         c.common.cdw10 = cpu_to_le32(cdw10);
2281
2282         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2283         nvme_put_ns_from_disk(head, srcu_idx);
2284         return ret;
2285 }
2286
2287 static int nvme_pr_register(struct block_device *bdev, u64 old,
2288                 u64 new, unsigned flags)
2289 {
2290         u32 cdw10;
2291
2292         if (flags & ~PR_FL_IGNORE_KEY)
2293                 return -EOPNOTSUPP;
2294
2295         cdw10 = old ? 2 : 0;
2296         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2297         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2298         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2299 }
2300
2301 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2302                 enum pr_type type, unsigned flags)
2303 {
2304         u32 cdw10;
2305
2306         if (flags & ~PR_FL_IGNORE_KEY)
2307                 return -EOPNOTSUPP;
2308
2309         cdw10 = nvme_pr_type(type) << 8;
2310         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2311         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2312 }
2313
2314 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2315                 enum pr_type type, bool abort)
2316 {
2317         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2318         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2319 }
2320
2321 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2322 {
2323         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2324         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2325 }
2326
2327 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2328 {
2329         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2330         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2331 }
2332
2333 static const struct pr_ops nvme_pr_ops = {
2334         .pr_register    = nvme_pr_register,
2335         .pr_reserve     = nvme_pr_reserve,
2336         .pr_release     = nvme_pr_release,
2337         .pr_preempt     = nvme_pr_preempt,
2338         .pr_clear       = nvme_pr_clear,
2339 };
2340
2341 #ifdef CONFIG_BLK_SED_OPAL
2342 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2343                 bool send)
2344 {
2345         struct nvme_ctrl *ctrl = data;
2346         struct nvme_command cmd;
2347
2348         memset(&cmd, 0, sizeof(cmd));
2349         if (send)
2350                 cmd.common.opcode = nvme_admin_security_send;
2351         else
2352                 cmd.common.opcode = nvme_admin_security_recv;
2353         cmd.common.nsid = 0;
2354         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2355         cmd.common.cdw11 = cpu_to_le32(len);
2356
2357         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2358                         NVME_QID_ANY, 1, 0, false);
2359 }
2360 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2361 #endif /* CONFIG_BLK_SED_OPAL */
2362
2363 static const struct block_device_operations nvme_bdev_ops = {
2364         .owner          = THIS_MODULE,
2365         .ioctl          = nvme_ioctl,
2366         .compat_ioctl   = nvme_compat_ioctl,
2367         .open           = nvme_open,
2368         .release        = nvme_release,
2369         .getgeo         = nvme_getgeo,
2370         .report_zones   = nvme_report_zones,
2371         .pr_ops         = &nvme_pr_ops,
2372 };
2373
2374 #ifdef CONFIG_NVME_MULTIPATH
2375 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2376 {
2377         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2378
2379         if (!kref_get_unless_zero(&head->ref))
2380                 return -ENXIO;
2381         return 0;
2382 }
2383
2384 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2385 {
2386         nvme_put_ns_head(disk->private_data);
2387 }
2388
2389 const struct block_device_operations nvme_ns_head_ops = {
2390         .owner          = THIS_MODULE,
2391         .submit_bio     = nvme_ns_head_submit_bio,
2392         .open           = nvme_ns_head_open,
2393         .release        = nvme_ns_head_release,
2394         .ioctl          = nvme_ioctl,
2395         .compat_ioctl   = nvme_compat_ioctl,
2396         .getgeo         = nvme_getgeo,
2397         .report_zones   = nvme_report_zones,
2398         .pr_ops         = &nvme_pr_ops,
2399 };
2400 #endif /* CONFIG_NVME_MULTIPATH */
2401
2402 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2403 {
2404         unsigned long timeout =
2405                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2406         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2407         int ret;
2408
2409         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2410                 if (csts == ~0)
2411                         return -ENODEV;
2412                 if ((csts & NVME_CSTS_RDY) == bit)
2413                         break;
2414
2415                 usleep_range(1000, 2000);
2416                 if (fatal_signal_pending(current))
2417                         return -EINTR;
2418                 if (time_after(jiffies, timeout)) {
2419                         dev_err(ctrl->device,
2420                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2421                                 enabled ? "initialisation" : "reset", csts);
2422                         return -ENODEV;
2423                 }
2424         }
2425
2426         return ret;
2427 }
2428
2429 /*
2430  * If the device has been passed off to us in an enabled state, just clear
2431  * the enabled bit.  The spec says we should set the 'shutdown notification
2432  * bits', but doing so may cause the device to complete commands to the
2433  * admin queue ... and we don't know what memory that might be pointing at!
2434  */
2435 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2436 {
2437         int ret;
2438
2439         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2440         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2441
2442         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2443         if (ret)
2444                 return ret;
2445
2446         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2447                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2448
2449         return nvme_wait_ready(ctrl, ctrl->cap, false);
2450 }
2451 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2452
2453 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2454 {
2455         unsigned dev_page_min;
2456         int ret;
2457
2458         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2459         if (ret) {
2460                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2461                 return ret;
2462         }
2463         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2464
2465         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2466                 dev_err(ctrl->device,
2467                         "Minimum device page size %u too large for host (%u)\n",
2468                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2469                 return -ENODEV;
2470         }
2471
2472         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2473                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2474         else
2475                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2476         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2477         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2478         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2479         ctrl->ctrl_config |= NVME_CC_ENABLE;
2480
2481         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2482         if (ret)
2483                 return ret;
2484         return nvme_wait_ready(ctrl, ctrl->cap, true);
2485 }
2486 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2487
2488 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2489 {
2490         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2491         u32 csts;
2492         int ret;
2493
2494         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2495         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2496
2497         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2498         if (ret)
2499                 return ret;
2500
2501         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2502                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2503                         break;
2504
2505                 msleep(100);
2506                 if (fatal_signal_pending(current))
2507                         return -EINTR;
2508                 if (time_after(jiffies, timeout)) {
2509                         dev_err(ctrl->device,
2510                                 "Device shutdown incomplete; abort shutdown\n");
2511                         return -ENODEV;
2512                 }
2513         }
2514
2515         return ret;
2516 }
2517 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2518
2519 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2520 {
2521         __le64 ts;
2522         int ret;
2523
2524         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2525                 return 0;
2526
2527         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2528         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2529                         NULL);
2530         if (ret)
2531                 dev_warn_once(ctrl->device,
2532                         "could not set timestamp (%d)\n", ret);
2533         return ret;
2534 }
2535
2536 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2537 {
2538         struct nvme_feat_host_behavior *host;
2539         int ret;
2540
2541         /* Don't bother enabling the feature if retry delay is not reported */
2542         if (!ctrl->crdt[0])
2543                 return 0;
2544
2545         host = kzalloc(sizeof(*host), GFP_KERNEL);
2546         if (!host)
2547                 return 0;
2548
2549         host->acre = NVME_ENABLE_ACRE;
2550         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2551                                 host, sizeof(*host), NULL);
2552         kfree(host);
2553         return ret;
2554 }
2555
2556 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2557 {
2558         /*
2559          * APST (Autonomous Power State Transition) lets us program a
2560          * table of power state transitions that the controller will
2561          * perform automatically.  We configure it with a simple
2562          * heuristic: we are willing to spend at most 2% of the time
2563          * transitioning between power states.  Therefore, when running
2564          * in any given state, we will enter the next lower-power
2565          * non-operational state after waiting 50 * (enlat + exlat)
2566          * microseconds, as long as that state's exit latency is under
2567          * the requested maximum latency.
2568          *
2569          * We will not autonomously enter any non-operational state for
2570          * which the total latency exceeds ps_max_latency_us.  Users
2571          * can set ps_max_latency_us to zero to turn off APST.
2572          */
2573
2574         unsigned apste;
2575         struct nvme_feat_auto_pst *table;
2576         u64 max_lat_us = 0;
2577         int max_ps = -1;
2578         int ret;
2579
2580         /*
2581          * If APST isn't supported or if we haven't been initialized yet,
2582          * then don't do anything.
2583          */
2584         if (!ctrl->apsta)
2585                 return 0;
2586
2587         if (ctrl->npss > 31) {
2588                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2589                 return 0;
2590         }
2591
2592         table = kzalloc(sizeof(*table), GFP_KERNEL);
2593         if (!table)
2594                 return 0;
2595
2596         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2597                 /* Turn off APST. */
2598                 apste = 0;
2599                 dev_dbg(ctrl->device, "APST disabled\n");
2600         } else {
2601                 __le64 target = cpu_to_le64(0);
2602                 int state;
2603
2604                 /*
2605                  * Walk through all states from lowest- to highest-power.
2606                  * According to the spec, lower-numbered states use more
2607                  * power.  NPSS, despite the name, is the index of the
2608                  * lowest-power state, not the number of states.
2609                  */
2610                 for (state = (int)ctrl->npss; state >= 0; state--) {
2611                         u64 total_latency_us, exit_latency_us, transition_ms;
2612
2613                         if (target)
2614                                 table->entries[state] = target;
2615
2616                         /*
2617                          * Don't allow transitions to the deepest state
2618                          * if it's quirked off.
2619                          */
2620                         if (state == ctrl->npss &&
2621                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2622                                 continue;
2623
2624                         /*
2625                          * Is this state a useful non-operational state for
2626                          * higher-power states to autonomously transition to?
2627                          */
2628                         if (!(ctrl->psd[state].flags &
2629                               NVME_PS_FLAGS_NON_OP_STATE))
2630                                 continue;
2631
2632                         exit_latency_us =
2633                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2634                         if (exit_latency_us > ctrl->ps_max_latency_us)
2635                                 continue;
2636
2637                         total_latency_us =
2638                                 exit_latency_us +
2639                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2640
2641                         /*
2642                          * This state is good.  Use it as the APST idle
2643                          * target for higher power states.
2644                          */
2645                         transition_ms = total_latency_us + 19;
2646                         do_div(transition_ms, 20);
2647                         if (transition_ms > (1 << 24) - 1)
2648                                 transition_ms = (1 << 24) - 1;
2649
2650                         target = cpu_to_le64((state << 3) |
2651                                              (transition_ms << 8));
2652
2653                         if (max_ps == -1)
2654                                 max_ps = state;
2655
2656                         if (total_latency_us > max_lat_us)
2657                                 max_lat_us = total_latency_us;
2658                 }
2659
2660                 apste = 1;
2661
2662                 if (max_ps == -1) {
2663                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2664                 } else {
2665                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2666                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2667                 }
2668         }
2669
2670         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2671                                 table, sizeof(*table), NULL);
2672         if (ret)
2673                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2674
2675         kfree(table);
2676         return ret;
2677 }
2678
2679 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2680 {
2681         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2682         u64 latency;
2683
2684         switch (val) {
2685         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2686         case PM_QOS_LATENCY_ANY:
2687                 latency = U64_MAX;
2688                 break;
2689
2690         default:
2691                 latency = val;
2692         }
2693
2694         if (ctrl->ps_max_latency_us != latency) {
2695                 ctrl->ps_max_latency_us = latency;
2696                 nvme_configure_apst(ctrl);
2697         }
2698 }
2699
2700 struct nvme_core_quirk_entry {
2701         /*
2702          * NVMe model and firmware strings are padded with spaces.  For
2703          * simplicity, strings in the quirk table are padded with NULLs
2704          * instead.
2705          */
2706         u16 vid;
2707         const char *mn;
2708         const char *fr;
2709         unsigned long quirks;
2710 };
2711
2712 static const struct nvme_core_quirk_entry core_quirks[] = {
2713         {
2714                 /*
2715                  * This Toshiba device seems to die using any APST states.  See:
2716                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2717                  */
2718                 .vid = 0x1179,
2719                 .mn = "THNSF5256GPUK TOSHIBA",
2720                 .quirks = NVME_QUIRK_NO_APST,
2721         },
2722         {
2723                 /*
2724                  * This LiteON CL1-3D*-Q11 firmware version has a race
2725                  * condition associated with actions related to suspend to idle
2726                  * LiteON has resolved the problem in future firmware
2727                  */
2728                 .vid = 0x14a4,
2729                 .fr = "22301111",
2730                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2731         }
2732 };
2733
2734 /* match is null-terminated but idstr is space-padded. */
2735 static bool string_matches(const char *idstr, const char *match, size_t len)
2736 {
2737         size_t matchlen;
2738
2739         if (!match)
2740                 return true;
2741
2742         matchlen = strlen(match);
2743         WARN_ON_ONCE(matchlen > len);
2744
2745         if (memcmp(idstr, match, matchlen))
2746                 return false;
2747
2748         for (; matchlen < len; matchlen++)
2749                 if (idstr[matchlen] != ' ')
2750                         return false;
2751
2752         return true;
2753 }
2754
2755 static bool quirk_matches(const struct nvme_id_ctrl *id,
2756                           const struct nvme_core_quirk_entry *q)
2757 {
2758         return q->vid == le16_to_cpu(id->vid) &&
2759                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2760                 string_matches(id->fr, q->fr, sizeof(id->fr));
2761 }
2762
2763 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2764                 struct nvme_id_ctrl *id)
2765 {
2766         size_t nqnlen;
2767         int off;
2768
2769         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2770                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2771                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2772                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2773                         return;
2774                 }
2775
2776                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2777                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2778         }
2779
2780         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2781         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2782                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2783                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2784         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2785         off += sizeof(id->sn);
2786         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2787         off += sizeof(id->mn);
2788         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2789 }
2790
2791 static void nvme_release_subsystem(struct device *dev)
2792 {
2793         struct nvme_subsystem *subsys =
2794                 container_of(dev, struct nvme_subsystem, dev);
2795
2796         if (subsys->instance >= 0)
2797                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2798         kfree(subsys);
2799 }
2800
2801 static void nvme_destroy_subsystem(struct kref *ref)
2802 {
2803         struct nvme_subsystem *subsys =
2804                         container_of(ref, struct nvme_subsystem, ref);
2805
2806         mutex_lock(&nvme_subsystems_lock);
2807         list_del(&subsys->entry);
2808         mutex_unlock(&nvme_subsystems_lock);
2809
2810         ida_destroy(&subsys->ns_ida);
2811         device_del(&subsys->dev);
2812         put_device(&subsys->dev);
2813 }
2814
2815 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2816 {
2817         kref_put(&subsys->ref, nvme_destroy_subsystem);
2818 }
2819
2820 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2821 {
2822         struct nvme_subsystem *subsys;
2823
2824         lockdep_assert_held(&nvme_subsystems_lock);
2825
2826         /*
2827          * Fail matches for discovery subsystems. This results
2828          * in each discovery controller bound to a unique subsystem.
2829          * This avoids issues with validating controller values
2830          * that can only be true when there is a single unique subsystem.
2831          * There may be multiple and completely independent entities
2832          * that provide discovery controllers.
2833          */
2834         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2835                 return NULL;
2836
2837         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2838                 if (strcmp(subsys->subnqn, subsysnqn))
2839                         continue;
2840                 if (!kref_get_unless_zero(&subsys->ref))
2841                         continue;
2842                 return subsys;
2843         }
2844
2845         return NULL;
2846 }
2847
2848 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2849         struct device_attribute subsys_attr_##_name = \
2850                 __ATTR(_name, _mode, _show, NULL)
2851
2852 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2853                                     struct device_attribute *attr,
2854                                     char *buf)
2855 {
2856         struct nvme_subsystem *subsys =
2857                 container_of(dev, struct nvme_subsystem, dev);
2858
2859         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2860 }
2861 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2862
2863 #define nvme_subsys_show_str_function(field)                            \
2864 static ssize_t subsys_##field##_show(struct device *dev,                \
2865                             struct device_attribute *attr, char *buf)   \
2866 {                                                                       \
2867         struct nvme_subsystem *subsys =                                 \
2868                 container_of(dev, struct nvme_subsystem, dev);          \
2869         return sprintf(buf, "%.*s\n",                                   \
2870                        (int)sizeof(subsys->field), subsys->field);      \
2871 }                                                                       \
2872 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2873
2874 nvme_subsys_show_str_function(model);
2875 nvme_subsys_show_str_function(serial);
2876 nvme_subsys_show_str_function(firmware_rev);
2877
2878 static struct attribute *nvme_subsys_attrs[] = {
2879         &subsys_attr_model.attr,
2880         &subsys_attr_serial.attr,
2881         &subsys_attr_firmware_rev.attr,
2882         &subsys_attr_subsysnqn.attr,
2883 #ifdef CONFIG_NVME_MULTIPATH
2884         &subsys_attr_iopolicy.attr,
2885 #endif
2886         NULL,
2887 };
2888
2889 static const struct attribute_group nvme_subsys_attrs_group = {
2890         .attrs = nvme_subsys_attrs,
2891 };
2892
2893 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2894         &nvme_subsys_attrs_group,
2895         NULL,
2896 };
2897
2898 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2899 {
2900         return ctrl->opts && ctrl->opts->discovery_nqn;
2901 }
2902
2903 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2904                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2905 {
2906         struct nvme_ctrl *tmp;
2907
2908         lockdep_assert_held(&nvme_subsystems_lock);
2909
2910         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2911                 if (nvme_state_terminal(tmp))
2912                         continue;
2913
2914                 if (tmp->cntlid == ctrl->cntlid) {
2915                         dev_err(ctrl->device,
2916                                 "Duplicate cntlid %u with %s, rejecting\n",
2917                                 ctrl->cntlid, dev_name(tmp->device));
2918                         return false;
2919                 }
2920
2921                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2922                     nvme_discovery_ctrl(ctrl))
2923                         continue;
2924
2925                 dev_err(ctrl->device,
2926                         "Subsystem does not support multiple controllers\n");
2927                 return false;
2928         }
2929
2930         return true;
2931 }
2932
2933 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2934 {
2935         struct nvme_subsystem *subsys, *found;
2936         int ret;
2937
2938         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2939         if (!subsys)
2940                 return -ENOMEM;
2941
2942         subsys->instance = -1;
2943         mutex_init(&subsys->lock);
2944         kref_init(&subsys->ref);
2945         INIT_LIST_HEAD(&subsys->ctrls);
2946         INIT_LIST_HEAD(&subsys->nsheads);
2947         nvme_init_subnqn(subsys, ctrl, id);
2948         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2949         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2950         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2951         subsys->vendor_id = le16_to_cpu(id->vid);
2952         subsys->cmic = id->cmic;
2953         subsys->awupf = le16_to_cpu(id->awupf);
2954 #ifdef CONFIG_NVME_MULTIPATH
2955         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2956 #endif
2957
2958         subsys->dev.class = nvme_subsys_class;
2959         subsys->dev.release = nvme_release_subsystem;
2960         subsys->dev.groups = nvme_subsys_attrs_groups;
2961         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2962         device_initialize(&subsys->dev);
2963
2964         mutex_lock(&nvme_subsystems_lock);
2965         found = __nvme_find_get_subsystem(subsys->subnqn);
2966         if (found) {
2967                 put_device(&subsys->dev);
2968                 subsys = found;
2969
2970                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2971                         ret = -EINVAL;
2972                         goto out_put_subsystem;
2973                 }
2974         } else {
2975                 ret = device_add(&subsys->dev);
2976                 if (ret) {
2977                         dev_err(ctrl->device,
2978                                 "failed to register subsystem device.\n");
2979                         put_device(&subsys->dev);
2980                         goto out_unlock;
2981                 }
2982                 ida_init(&subsys->ns_ida);
2983                 list_add_tail(&subsys->entry, &nvme_subsystems);
2984         }
2985
2986         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2987                                 dev_name(ctrl->device));
2988         if (ret) {
2989                 dev_err(ctrl->device,
2990                         "failed to create sysfs link from subsystem.\n");
2991                 goto out_put_subsystem;
2992         }
2993
2994         if (!found)
2995                 subsys->instance = ctrl->instance;
2996         ctrl->subsys = subsys;
2997         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2998         mutex_unlock(&nvme_subsystems_lock);
2999         return 0;
3000
3001 out_put_subsystem:
3002         nvme_put_subsystem(subsys);
3003 out_unlock:
3004         mutex_unlock(&nvme_subsystems_lock);
3005         return ret;
3006 }
3007
3008 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3009                 void *log, size_t size, u64 offset)
3010 {
3011         struct nvme_command c = { };
3012         u32 dwlen = nvme_bytes_to_numd(size);
3013
3014         c.get_log_page.opcode = nvme_admin_get_log_page;
3015         c.get_log_page.nsid = cpu_to_le32(nsid);
3016         c.get_log_page.lid = log_page;
3017         c.get_log_page.lsp = lsp;
3018         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3019         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3020         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3021         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3022         c.get_log_page.csi = csi;
3023
3024         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3025 }
3026
3027 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3028                                 struct nvme_effects_log **log)
3029 {
3030         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3031         int ret;
3032
3033         if (cel)
3034                 goto out;
3035
3036         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3037         if (!cel)
3038                 return -ENOMEM;
3039
3040         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3041                         cel, sizeof(*cel), 0);
3042         if (ret) {
3043                 kfree(cel);
3044                 return ret;
3045         }
3046
3047         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3048 out:
3049         *log = cel;
3050         return 0;
3051 }
3052
3053 /*
3054  * Initialize the cached copies of the Identify data and various controller
3055  * register in our nvme_ctrl structure.  This should be called as soon as
3056  * the admin queue is fully up and running.
3057  */
3058 int nvme_init_identify(struct nvme_ctrl *ctrl)
3059 {
3060         struct nvme_id_ctrl *id;
3061         int ret, page_shift;
3062         u32 max_hw_sectors;
3063         bool prev_apst_enabled;
3064
3065         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3066         if (ret) {
3067                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3068                 return ret;
3069         }
3070         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3071         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3072
3073         if (ctrl->vs >= NVME_VS(1, 1, 0))
3074                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3075
3076         ret = nvme_identify_ctrl(ctrl, &id);
3077         if (ret) {
3078                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3079                 return -EIO;
3080         }
3081
3082         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3083                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3084                 if (ret < 0)
3085                         goto out_free;
3086         }
3087
3088         if (!(ctrl->ops->flags & NVME_F_FABRICS))
3089                 ctrl->cntlid = le16_to_cpu(id->cntlid);
3090
3091         if (!ctrl->identified) {
3092                 int i;
3093
3094                 ret = nvme_init_subsystem(ctrl, id);
3095                 if (ret)
3096                         goto out_free;
3097
3098                 /*
3099                  * Check for quirks.  Quirk can depend on firmware version,
3100                  * so, in principle, the set of quirks present can change
3101                  * across a reset.  As a possible future enhancement, we
3102                  * could re-scan for quirks every time we reinitialize
3103                  * the device, but we'd have to make sure that the driver
3104                  * behaves intelligently if the quirks change.
3105                  */
3106                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3107                         if (quirk_matches(id, &core_quirks[i]))
3108                                 ctrl->quirks |= core_quirks[i].quirks;
3109                 }
3110         }
3111
3112         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3113                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3114                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3115         }
3116
3117         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3118         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3119         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3120
3121         ctrl->oacs = le16_to_cpu(id->oacs);
3122         ctrl->oncs = le16_to_cpu(id->oncs);
3123         ctrl->mtfa = le16_to_cpu(id->mtfa);
3124         ctrl->oaes = le32_to_cpu(id->oaes);
3125         ctrl->wctemp = le16_to_cpu(id->wctemp);
3126         ctrl->cctemp = le16_to_cpu(id->cctemp);
3127
3128         atomic_set(&ctrl->abort_limit, id->acl + 1);
3129         ctrl->vwc = id->vwc;
3130         if (id->mdts)
3131                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3132         else
3133                 max_hw_sectors = UINT_MAX;
3134         ctrl->max_hw_sectors =
3135                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3136
3137         nvme_set_queue_limits(ctrl, ctrl->admin_q);
3138         ctrl->sgls = le32_to_cpu(id->sgls);
3139         ctrl->kas = le16_to_cpu(id->kas);
3140         ctrl->max_namespaces = le32_to_cpu(id->mnan);
3141         ctrl->ctratt = le32_to_cpu(id->ctratt);
3142
3143         if (id->rtd3e) {
3144                 /* us -> s */
3145                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3146
3147                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3148                                                  shutdown_timeout, 60);
3149
3150                 if (ctrl->shutdown_timeout != shutdown_timeout)
3151                         dev_info(ctrl->device,
3152                                  "Shutdown timeout set to %u seconds\n",
3153                                  ctrl->shutdown_timeout);
3154         } else
3155                 ctrl->shutdown_timeout = shutdown_timeout;
3156
3157         ctrl->npss = id->npss;
3158         ctrl->apsta = id->apsta;
3159         prev_apst_enabled = ctrl->apst_enabled;
3160         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3161                 if (force_apst && id->apsta) {
3162                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3163                         ctrl->apst_enabled = true;
3164                 } else {
3165                         ctrl->apst_enabled = false;
3166                 }
3167         } else {
3168                 ctrl->apst_enabled = id->apsta;
3169         }
3170         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3171
3172         if (ctrl->ops->flags & NVME_F_FABRICS) {
3173                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3174                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3175                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3176                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3177
3178                 /*
3179                  * In fabrics we need to verify the cntlid matches the
3180                  * admin connect
3181                  */
3182                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3183                         dev_err(ctrl->device,
3184                                 "Mismatching cntlid: Connect %u vs Identify "
3185                                 "%u, rejecting\n",
3186                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3187                         ret = -EINVAL;
3188                         goto out_free;
3189                 }
3190
3191                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3192                         dev_err(ctrl->device,
3193                                 "keep-alive support is mandatory for fabrics\n");
3194                         ret = -EINVAL;
3195                         goto out_free;
3196                 }
3197         } else {
3198                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3199                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3200                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3201                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3202         }
3203
3204         ret = nvme_mpath_init(ctrl, id);
3205         kfree(id);
3206
3207         if (ret < 0)
3208                 return ret;
3209
3210         if (ctrl->apst_enabled && !prev_apst_enabled)
3211                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3212         else if (!ctrl->apst_enabled && prev_apst_enabled)
3213                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3214
3215         ret = nvme_configure_apst(ctrl);
3216         if (ret < 0)
3217                 return ret;
3218
3219         ret = nvme_configure_timestamp(ctrl);
3220         if (ret < 0)
3221                 return ret;
3222
3223         ret = nvme_configure_directives(ctrl);
3224         if (ret < 0)
3225                 return ret;
3226
3227         ret = nvme_configure_acre(ctrl);
3228         if (ret < 0)
3229                 return ret;
3230
3231         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3232                 ret = nvme_hwmon_init(ctrl);
3233                 if (ret < 0)
3234                         return ret;
3235         }
3236
3237         ctrl->identified = true;
3238
3239         return 0;
3240
3241 out_free:
3242         kfree(id);
3243         return ret;
3244 }
3245 EXPORT_SYMBOL_GPL(nvme_init_identify);
3246
3247 static int nvme_dev_open(struct inode *inode, struct file *file)
3248 {
3249         struct nvme_ctrl *ctrl =
3250                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3251
3252         switch (ctrl->state) {
3253         case NVME_CTRL_LIVE:
3254                 break;
3255         default:
3256                 return -EWOULDBLOCK;
3257         }
3258
3259         nvme_get_ctrl(ctrl);
3260         if (!try_module_get(ctrl->ops->module)) {
3261                 nvme_put_ctrl(ctrl);
3262                 return -EINVAL;
3263         }
3264
3265         file->private_data = ctrl;
3266         return 0;
3267 }
3268
3269 static int nvme_dev_release(struct inode *inode, struct file *file)
3270 {
3271         struct nvme_ctrl *ctrl =
3272                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3273
3274         module_put(ctrl->ops->module);
3275         nvme_put_ctrl(ctrl);
3276         return 0;
3277 }
3278
3279 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3280 {
3281         struct nvme_ns *ns;
3282         int ret;
3283
3284         down_read(&ctrl->namespaces_rwsem);
3285         if (list_empty(&ctrl->namespaces)) {
3286                 ret = -ENOTTY;
3287                 goto out_unlock;
3288         }
3289
3290         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3291         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3292                 dev_warn(ctrl->device,
3293                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3294                 ret = -EINVAL;
3295                 goto out_unlock;
3296         }
3297
3298         dev_warn(ctrl->device,
3299                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3300         kref_get(&ns->kref);
3301         up_read(&ctrl->namespaces_rwsem);
3302
3303         ret = nvme_user_cmd(ctrl, ns, argp);
3304         nvme_put_ns(ns);
3305         return ret;
3306
3307 out_unlock:
3308         up_read(&ctrl->namespaces_rwsem);
3309         return ret;
3310 }
3311
3312 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3313                 unsigned long arg)
3314 {
3315         struct nvme_ctrl *ctrl = file->private_data;
3316         void __user *argp = (void __user *)arg;
3317
3318         switch (cmd) {
3319         case NVME_IOCTL_ADMIN_CMD:
3320                 return nvme_user_cmd(ctrl, NULL, argp);
3321         case NVME_IOCTL_ADMIN64_CMD:
3322                 return nvme_user_cmd64(ctrl, NULL, argp);
3323         case NVME_IOCTL_IO_CMD:
3324                 return nvme_dev_user_cmd(ctrl, argp);
3325         case NVME_IOCTL_RESET:
3326                 dev_warn(ctrl->device, "resetting controller\n");
3327                 return nvme_reset_ctrl_sync(ctrl);
3328         case NVME_IOCTL_SUBSYS_RESET:
3329                 return nvme_reset_subsystem(ctrl);
3330         case NVME_IOCTL_RESCAN:
3331                 nvme_queue_scan(ctrl);
3332                 return 0;
3333         default:
3334                 return -ENOTTY;
3335         }
3336 }
3337
3338 static const struct file_operations nvme_dev_fops = {
3339         .owner          = THIS_MODULE,
3340         .open           = nvme_dev_open,
3341         .release        = nvme_dev_release,
3342         .unlocked_ioctl = nvme_dev_ioctl,
3343         .compat_ioctl   = compat_ptr_ioctl,
3344 };
3345
3346 static ssize_t nvme_sysfs_reset(struct device *dev,
3347                                 struct device_attribute *attr, const char *buf,
3348                                 size_t count)
3349 {
3350         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3351         int ret;
3352
3353         ret = nvme_reset_ctrl_sync(ctrl);
3354         if (ret < 0)
3355                 return ret;
3356         return count;
3357 }
3358 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3359
3360 static ssize_t nvme_sysfs_rescan(struct device *dev,
3361                                 struct device_attribute *attr, const char *buf,
3362                                 size_t count)
3363 {
3364         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3365
3366         nvme_queue_scan(ctrl);
3367         return count;
3368 }
3369 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3370
3371 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3372 {
3373         struct gendisk *disk = dev_to_disk(dev);
3374
3375         if (disk->fops == &nvme_bdev_ops)
3376                 return nvme_get_ns_from_dev(dev)->head;
3377         else
3378                 return disk->private_data;
3379 }
3380
3381 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3382                 char *buf)
3383 {
3384         struct nvme_ns_head *head = dev_to_ns_head(dev);
3385         struct nvme_ns_ids *ids = &head->ids;
3386         struct nvme_subsystem *subsys = head->subsys;
3387         int serial_len = sizeof(subsys->serial);
3388         int model_len = sizeof(subsys->model);
3389
3390         if (!uuid_is_null(&ids->uuid))
3391                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3392
3393         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3394                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3395
3396         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3397                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3398
3399         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3400                                   subsys->serial[serial_len - 1] == '\0'))
3401                 serial_len--;
3402         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3403                                  subsys->model[model_len - 1] == '\0'))
3404                 model_len--;
3405
3406         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3407                 serial_len, subsys->serial, model_len, subsys->model,
3408                 head->ns_id);
3409 }
3410 static DEVICE_ATTR_RO(wwid);
3411
3412 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3413                 char *buf)
3414 {
3415         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3416 }
3417 static DEVICE_ATTR_RO(nguid);
3418
3419 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3420                 char *buf)
3421 {
3422         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3423
3424         /* For backward compatibility expose the NGUID to userspace if
3425          * we have no UUID set
3426          */
3427         if (uuid_is_null(&ids->uuid)) {
3428                 printk_ratelimited(KERN_WARNING
3429                                    "No UUID available providing old NGUID\n");
3430                 return sprintf(buf, "%pU\n", ids->nguid);
3431         }
3432         return sprintf(buf, "%pU\n", &ids->uuid);
3433 }
3434 static DEVICE_ATTR_RO(uuid);
3435
3436 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3437                 char *buf)
3438 {
3439         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3440 }
3441 static DEVICE_ATTR_RO(eui);
3442
3443 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3444                 char *buf)
3445 {
3446         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3447 }
3448 static DEVICE_ATTR_RO(nsid);
3449
3450 static struct attribute *nvme_ns_id_attrs[] = {
3451         &dev_attr_wwid.attr,
3452         &dev_attr_uuid.attr,
3453         &dev_attr_nguid.attr,
3454         &dev_attr_eui.attr,
3455         &dev_attr_nsid.attr,
3456 #ifdef CONFIG_NVME_MULTIPATH
3457         &dev_attr_ana_grpid.attr,
3458         &dev_attr_ana_state.attr,
3459 #endif
3460         NULL,
3461 };
3462
3463 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3464                 struct attribute *a, int n)
3465 {
3466         struct device *dev = container_of(kobj, struct device, kobj);
3467         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3468
3469         if (a == &dev_attr_uuid.attr) {
3470                 if (uuid_is_null(&ids->uuid) &&
3471                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3472                         return 0;
3473         }
3474         if (a == &dev_attr_nguid.attr) {
3475                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3476                         return 0;
3477         }
3478         if (a == &dev_attr_eui.attr) {
3479                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3480                         return 0;
3481         }
3482 #ifdef CONFIG_NVME_MULTIPATH
3483         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3484                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3485                         return 0;
3486                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3487                         return 0;
3488         }
3489 #endif
3490         return a->mode;
3491 }
3492
3493 static const struct attribute_group nvme_ns_id_attr_group = {
3494         .attrs          = nvme_ns_id_attrs,
3495         .is_visible     = nvme_ns_id_attrs_are_visible,
3496 };
3497
3498 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3499         &nvme_ns_id_attr_group,
3500 #ifdef CONFIG_NVM
3501         &nvme_nvm_attr_group,
3502 #endif
3503         NULL,
3504 };
3505
3506 #define nvme_show_str_function(field)                                           \
3507 static ssize_t  field##_show(struct device *dev,                                \
3508                             struct device_attribute *attr, char *buf)           \
3509 {                                                                               \
3510         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3511         return sprintf(buf, "%.*s\n",                                           \
3512                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3513 }                                                                               \
3514 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3515
3516 nvme_show_str_function(model);
3517 nvme_show_str_function(serial);
3518 nvme_show_str_function(firmware_rev);
3519
3520 #define nvme_show_int_function(field)                                           \
3521 static ssize_t  field##_show(struct device *dev,                                \
3522                             struct device_attribute *attr, char *buf)           \
3523 {                                                                               \
3524         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3525         return sprintf(buf, "%d\n", ctrl->field);       \
3526 }                                                                               \
3527 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3528
3529 nvme_show_int_function(cntlid);
3530 nvme_show_int_function(numa_node);
3531 nvme_show_int_function(queue_count);
3532 nvme_show_int_function(sqsize);
3533
3534 static ssize_t nvme_sysfs_delete(struct device *dev,
3535                                 struct device_attribute *attr, const char *buf,
3536                                 size_t count)
3537 {
3538         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3539
3540         if (device_remove_file_self(dev, attr))
3541                 nvme_delete_ctrl_sync(ctrl);
3542         return count;
3543 }
3544 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3545
3546 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3547                                          struct device_attribute *attr,
3548                                          char *buf)
3549 {
3550         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3551
3552         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3553 }
3554 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3555
3556 static ssize_t nvme_sysfs_show_state(struct device *dev,
3557                                      struct device_attribute *attr,
3558                                      char *buf)
3559 {
3560         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3561         static const char *const state_name[] = {
3562                 [NVME_CTRL_NEW]         = "new",
3563                 [NVME_CTRL_LIVE]        = "live",
3564                 [NVME_CTRL_RESETTING]   = "resetting",
3565                 [NVME_CTRL_CONNECTING]  = "connecting",
3566                 [NVME_CTRL_DELETING]    = "deleting",
3567                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3568                 [NVME_CTRL_DEAD]        = "dead",
3569         };
3570
3571         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3572             state_name[ctrl->state])
3573                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3574
3575         return sprintf(buf, "unknown state\n");
3576 }
3577
3578 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3579
3580 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3581                                          struct device_attribute *attr,
3582                                          char *buf)
3583 {
3584         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3585
3586         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3587 }
3588 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3589
3590 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3591                                         struct device_attribute *attr,
3592                                         char *buf)
3593 {
3594         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3595
3596         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3597 }
3598 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3599
3600 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3601                                         struct device_attribute *attr,
3602                                         char *buf)
3603 {
3604         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3605
3606         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3607 }
3608 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3609
3610 static ssize_t nvme_sysfs_show_address(struct device *dev,
3611                                          struct device_attribute *attr,
3612                                          char *buf)
3613 {
3614         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3615
3616         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3617 }
3618 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3619
3620 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3621                 struct device_attribute *attr, char *buf)
3622 {
3623         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3624         struct nvmf_ctrl_options *opts = ctrl->opts;
3625
3626         if (ctrl->opts->max_reconnects == -1)
3627                 return sprintf(buf, "off\n");
3628         return sprintf(buf, "%d\n",
3629                         opts->max_reconnects * opts->reconnect_delay);
3630 }
3631
3632 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3633                 struct device_attribute *attr, const char *buf, size_t count)
3634 {
3635         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3636         struct nvmf_ctrl_options *opts = ctrl->opts;
3637         int ctrl_loss_tmo, err;
3638
3639         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3640         if (err)
3641                 return -EINVAL;
3642
3643         else if (ctrl_loss_tmo < 0)
3644                 opts->max_reconnects = -1;
3645         else
3646                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3647                                                 opts->reconnect_delay);
3648         return count;
3649 }
3650 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3651         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3652
3653 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3654                 struct device_attribute *attr, char *buf)
3655 {
3656         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3657
3658         if (ctrl->opts->reconnect_delay == -1)
3659                 return sprintf(buf, "off\n");
3660         return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3661 }
3662
3663 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3664                 struct device_attribute *attr, const char *buf, size_t count)
3665 {
3666         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3667         unsigned int v;
3668         int err;
3669
3670         err = kstrtou32(buf, 10, &v);
3671         if (err)
3672                 return err;
3673
3674         ctrl->opts->reconnect_delay = v;
3675         return count;
3676 }
3677 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3678         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3679
3680 static struct attribute *nvme_dev_attrs[] = {
3681         &dev_attr_reset_controller.attr,
3682         &dev_attr_rescan_controller.attr,
3683         &dev_attr_model.attr,
3684         &dev_attr_serial.attr,
3685         &dev_attr_firmware_rev.attr,
3686         &dev_attr_cntlid.attr,
3687         &dev_attr_delete_controller.attr,
3688         &dev_attr_transport.attr,
3689         &dev_attr_subsysnqn.attr,
3690         &dev_attr_address.attr,
3691         &dev_attr_state.attr,
3692         &dev_attr_numa_node.attr,
3693         &dev_attr_queue_count.attr,
3694         &dev_attr_sqsize.attr,
3695         &dev_attr_hostnqn.attr,
3696         &dev_attr_hostid.attr,
3697         &dev_attr_ctrl_loss_tmo.attr,
3698         &dev_attr_reconnect_delay.attr,
3699         NULL
3700 };
3701
3702 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3703                 struct attribute *a, int n)
3704 {
3705         struct device *dev = container_of(kobj, struct device, kobj);
3706         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3707
3708         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3709                 return 0;
3710         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3711                 return 0;
3712         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3713                 return 0;
3714         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3715                 return 0;
3716         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3717                 return 0;
3718         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3719                 return 0;
3720
3721         return a->mode;
3722 }
3723
3724 static const struct attribute_group nvme_dev_attrs_group = {
3725         .attrs          = nvme_dev_attrs,
3726         .is_visible     = nvme_dev_attrs_are_visible,
3727 };
3728
3729 static const struct attribute_group *nvme_dev_attr_groups[] = {
3730         &nvme_dev_attrs_group,
3731         NULL,
3732 };
3733
3734 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3735                 unsigned nsid)
3736 {
3737         struct nvme_ns_head *h;
3738
3739         lockdep_assert_held(&subsys->lock);
3740
3741         list_for_each_entry(h, &subsys->nsheads, entry) {
3742                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3743                         return h;
3744         }
3745
3746         return NULL;
3747 }
3748
3749 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3750                 struct nvme_ns_head *new)
3751 {
3752         struct nvme_ns_head *h;
3753
3754         lockdep_assert_held(&subsys->lock);
3755
3756         list_for_each_entry(h, &subsys->nsheads, entry) {
3757                 if (nvme_ns_ids_valid(&new->ids) &&
3758                     nvme_ns_ids_equal(&new->ids, &h->ids))
3759                         return -EINVAL;
3760         }
3761
3762         return 0;
3763 }
3764
3765 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3766                 unsigned nsid, struct nvme_ns_ids *ids)
3767 {
3768         struct nvme_ns_head *head;
3769         size_t size = sizeof(*head);
3770         int ret = -ENOMEM;
3771
3772 #ifdef CONFIG_NVME_MULTIPATH
3773         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3774 #endif
3775
3776         head = kzalloc(size, GFP_KERNEL);
3777         if (!head)
3778                 goto out;
3779         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3780         if (ret < 0)
3781                 goto out_free_head;
3782         head->instance = ret;
3783         INIT_LIST_HEAD(&head->list);
3784         ret = init_srcu_struct(&head->srcu);
3785         if (ret)
3786                 goto out_ida_remove;
3787         head->subsys = ctrl->subsys;
3788         head->ns_id = nsid;
3789         head->ids = *ids;
3790         kref_init(&head->ref);
3791
3792         ret = __nvme_check_ids(ctrl->subsys, head);
3793         if (ret) {
3794                 dev_err(ctrl->device,
3795                         "duplicate IDs for nsid %d\n", nsid);
3796                 goto out_cleanup_srcu;
3797         }
3798
3799         if (head->ids.csi) {
3800                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3801                 if (ret)
3802                         goto out_cleanup_srcu;
3803         } else
3804                 head->effects = ctrl->effects;
3805
3806         ret = nvme_mpath_alloc_disk(ctrl, head);
3807         if (ret)
3808                 goto out_cleanup_srcu;
3809
3810         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3811
3812         kref_get(&ctrl->subsys->ref);
3813
3814         return head;
3815 out_cleanup_srcu:
3816         cleanup_srcu_struct(&head->srcu);
3817 out_ida_remove:
3818         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3819 out_free_head:
3820         kfree(head);
3821 out:
3822         if (ret > 0)
3823                 ret = blk_status_to_errno(nvme_error_status(ret));
3824         return ERR_PTR(ret);
3825 }
3826
3827 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3828                 struct nvme_ns_ids *ids, bool is_shared)
3829 {
3830         struct nvme_ctrl *ctrl = ns->ctrl;
3831         struct nvme_ns_head *head = NULL;
3832         int ret = 0;
3833
3834         mutex_lock(&ctrl->subsys->lock);
3835         head = nvme_find_ns_head(ctrl->subsys, nsid);
3836         if (!head) {
3837                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3838                 if (IS_ERR(head)) {
3839                         ret = PTR_ERR(head);
3840                         goto out_unlock;
3841                 }
3842                 head->shared = is_shared;
3843         } else {
3844                 ret = -EINVAL;
3845                 if (!is_shared || !head->shared) {
3846                         dev_err(ctrl->device,
3847                                 "Duplicate unshared namespace %d\n", nsid);
3848                         goto out_put_ns_head;
3849                 }
3850                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3851                         dev_err(ctrl->device,
3852                                 "IDs don't match for shared namespace %d\n",
3853                                         nsid);
3854                         goto out_put_ns_head;
3855                 }
3856         }
3857
3858         list_add_tail_rcu(&ns->siblings, &head->list);
3859         ns->head = head;
3860         mutex_unlock(&ctrl->subsys->lock);
3861         return 0;
3862
3863 out_put_ns_head:
3864         nvme_put_ns_head(head);
3865 out_unlock:
3866         mutex_unlock(&ctrl->subsys->lock);
3867         return ret;
3868 }
3869
3870 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3871 {
3872         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3873         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3874
3875         return nsa->head->ns_id - nsb->head->ns_id;
3876 }
3877
3878 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3879 {
3880         struct nvme_ns *ns, *ret = NULL;
3881
3882         down_read(&ctrl->namespaces_rwsem);
3883         list_for_each_entry(ns, &ctrl->namespaces, list) {
3884                 if (ns->head->ns_id == nsid) {
3885                         if (!kref_get_unless_zero(&ns->kref))
3886                                 continue;
3887                         ret = ns;
3888                         break;
3889                 }
3890                 if (ns->head->ns_id > nsid)
3891                         break;
3892         }
3893         up_read(&ctrl->namespaces_rwsem);
3894         return ret;
3895 }
3896 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3897
3898 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3899                 struct nvme_ns_ids *ids)
3900 {
3901         struct nvme_ns *ns;
3902         struct gendisk *disk;
3903         struct nvme_id_ns *id;
3904         char disk_name[DISK_NAME_LEN];
3905         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3906
3907         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3908                 return;
3909
3910         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3911         if (!ns)
3912                 goto out_free_id;
3913
3914         ns->queue = blk_mq_init_queue(ctrl->tagset);
3915         if (IS_ERR(ns->queue))
3916                 goto out_free_ns;
3917
3918         if (ctrl->opts && ctrl->opts->data_digest)
3919                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3920
3921         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3922         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3923                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3924
3925         ns->queue->queuedata = ns;
3926         ns->ctrl = ctrl;
3927         kref_init(&ns->kref);
3928
3929         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3930                 goto out_free_queue;
3931         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3932
3933         disk = alloc_disk_node(0, node);
3934         if (!disk)
3935                 goto out_unlink_ns;
3936
3937         disk->fops = &nvme_bdev_ops;
3938         disk->private_data = ns;
3939         disk->queue = ns->queue;
3940         disk->flags = flags;
3941         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3942         ns->disk = disk;
3943
3944         if (nvme_update_ns_info(ns, id))
3945                 goto out_put_disk;
3946
3947         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3948                 if (nvme_nvm_register(ns, disk_name, node)) {
3949                         dev_warn(ctrl->device, "LightNVM init failure\n");
3950                         goto out_put_disk;
3951                 }
3952         }
3953
3954         down_write(&ctrl->namespaces_rwsem);
3955         list_add_tail(&ns->list, &ctrl->namespaces);
3956         up_write(&ctrl->namespaces_rwsem);
3957
3958         nvme_get_ctrl(ctrl);
3959
3960         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3961
3962         nvme_mpath_add_disk(ns, id);
3963         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3964         kfree(id);
3965
3966         return;
3967  out_put_disk:
3968         /* prevent double queue cleanup */
3969         ns->disk->queue = NULL;
3970         put_disk(ns->disk);
3971  out_unlink_ns:
3972         mutex_lock(&ctrl->subsys->lock);
3973         list_del_rcu(&ns->siblings);
3974         if (list_empty(&ns->head->list))
3975                 list_del_init(&ns->head->entry);
3976         mutex_unlock(&ctrl->subsys->lock);
3977         nvme_put_ns_head(ns->head);
3978  out_free_queue:
3979         blk_cleanup_queue(ns->queue);
3980  out_free_ns:
3981         kfree(ns);
3982  out_free_id:
3983         kfree(id);
3984 }
3985
3986 static void nvme_ns_remove(struct nvme_ns *ns)
3987 {
3988         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3989                 return;
3990
3991         set_capacity(ns->disk, 0);
3992         nvme_fault_inject_fini(&ns->fault_inject);
3993
3994         mutex_lock(&ns->ctrl->subsys->lock);
3995         list_del_rcu(&ns->siblings);
3996         if (list_empty(&ns->head->list))
3997                 list_del_init(&ns->head->entry);
3998         mutex_unlock(&ns->ctrl->subsys->lock);
3999
4000         synchronize_rcu(); /* guarantee not available in head->list */
4001         nvme_mpath_clear_current_path(ns);
4002         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4003
4004         if (ns->disk->flags & GENHD_FL_UP) {
4005                 del_gendisk(ns->disk);
4006                 blk_cleanup_queue(ns->queue);
4007                 if (blk_get_integrity(ns->disk))
4008                         blk_integrity_unregister(ns->disk);
4009         }
4010
4011         down_write(&ns->ctrl->namespaces_rwsem);
4012         list_del_init(&ns->list);
4013         up_write(&ns->ctrl->namespaces_rwsem);
4014
4015         nvme_mpath_check_last_path(ns);
4016         nvme_put_ns(ns);
4017 }
4018
4019 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4020 {
4021         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4022
4023         if (ns) {
4024                 nvme_ns_remove(ns);
4025                 nvme_put_ns(ns);
4026         }
4027 }
4028
4029 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4030 {
4031         struct nvme_id_ns *id;
4032         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4033
4034         if (test_bit(NVME_NS_DEAD, &ns->flags))
4035                 goto out;
4036
4037         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4038         if (ret)
4039                 goto out;
4040
4041         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4042         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4043                 dev_err(ns->ctrl->device,
4044                         "identifiers changed for nsid %d\n", ns->head->ns_id);
4045                 goto out_free_id;
4046         }
4047
4048         ret = nvme_update_ns_info(ns, id);
4049
4050 out_free_id:
4051         kfree(id);
4052 out:
4053         /*
4054          * Only remove the namespace if we got a fatal error back from the
4055          * device, otherwise ignore the error and just move on.
4056          *
4057          * TODO: we should probably schedule a delayed retry here.
4058          */
4059         if (ret > 0 && (ret & NVME_SC_DNR))
4060                 nvme_ns_remove(ns);
4061 }
4062
4063 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4064 {
4065         struct nvme_ns_ids ids = { };
4066         struct nvme_ns *ns;
4067
4068         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4069                 return;
4070
4071         ns = nvme_find_get_ns(ctrl, nsid);
4072         if (ns) {
4073                 nvme_validate_ns(ns, &ids);
4074                 nvme_put_ns(ns);
4075                 return;
4076         }
4077
4078         switch (ids.csi) {
4079         case NVME_CSI_NVM:
4080                 nvme_alloc_ns(ctrl, nsid, &ids);
4081                 break;
4082         case NVME_CSI_ZNS:
4083                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4084                         dev_warn(ctrl->device,
4085                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4086                                 nsid);
4087                         break;
4088                 }
4089                 if (!nvme_multi_css(ctrl)) {
4090                         dev_warn(ctrl->device,
4091                                 "command set not reported for nsid: %d\n",
4092                                 nsid);
4093                         break;
4094                 }
4095                 nvme_alloc_ns(ctrl, nsid, &ids);
4096                 break;
4097         default:
4098                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4099                         ids.csi, nsid);
4100                 break;
4101         }
4102 }
4103
4104 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4105                                         unsigned nsid)
4106 {
4107         struct nvme_ns *ns, *next;
4108         LIST_HEAD(rm_list);
4109
4110         down_write(&ctrl->namespaces_rwsem);
4111         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4112                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4113                         list_move_tail(&ns->list, &rm_list);
4114         }
4115         up_write(&ctrl->namespaces_rwsem);
4116
4117         list_for_each_entry_safe(ns, next, &rm_list, list)
4118                 nvme_ns_remove(ns);
4119
4120 }
4121
4122 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4123 {
4124         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4125         __le32 *ns_list;
4126         u32 prev = 0;
4127         int ret = 0, i;
4128
4129         if (nvme_ctrl_limited_cns(ctrl))
4130                 return -EOPNOTSUPP;
4131
4132         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4133         if (!ns_list)
4134                 return -ENOMEM;
4135
4136         for (;;) {
4137                 struct nvme_command cmd = {
4138                         .identify.opcode        = nvme_admin_identify,
4139                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4140                         .identify.nsid          = cpu_to_le32(prev),
4141                 };
4142
4143                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4144                                             NVME_IDENTIFY_DATA_SIZE);
4145                 if (ret) {
4146                         dev_warn(ctrl->device,
4147                                 "Identify NS List failed (status=0x%x)\n", ret);
4148                         goto free;
4149                 }
4150
4151                 for (i = 0; i < nr_entries; i++) {
4152                         u32 nsid = le32_to_cpu(ns_list[i]);
4153
4154                         if (!nsid)      /* end of the list? */
4155                                 goto out;
4156                         nvme_validate_or_alloc_ns(ctrl, nsid);
4157                         while (++prev < nsid)
4158                                 nvme_ns_remove_by_nsid(ctrl, prev);
4159                 }
4160         }
4161  out:
4162         nvme_remove_invalid_namespaces(ctrl, prev);
4163  free:
4164         kfree(ns_list);
4165         return ret;
4166 }
4167
4168 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4169 {
4170         struct nvme_id_ctrl *id;
4171         u32 nn, i;
4172
4173         if (nvme_identify_ctrl(ctrl, &id))
4174                 return;
4175         nn = le32_to_cpu(id->nn);
4176         kfree(id);
4177
4178         for (i = 1; i <= nn; i++)
4179                 nvme_validate_or_alloc_ns(ctrl, i);
4180
4181         nvme_remove_invalid_namespaces(ctrl, nn);
4182 }
4183
4184 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4185 {
4186         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4187         __le32 *log;
4188         int error;
4189
4190         log = kzalloc(log_size, GFP_KERNEL);
4191         if (!log)
4192                 return;
4193
4194         /*
4195          * We need to read the log to clear the AEN, but we don't want to rely
4196          * on it for the changed namespace information as userspace could have
4197          * raced with us in reading the log page, which could cause us to miss
4198          * updates.
4199          */
4200         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4201                         NVME_CSI_NVM, log, log_size, 0);
4202         if (error)
4203                 dev_warn(ctrl->device,
4204                         "reading changed ns log failed: %d\n", error);
4205
4206         kfree(log);
4207 }
4208
4209 static void nvme_scan_work(struct work_struct *work)
4210 {
4211         struct nvme_ctrl *ctrl =
4212                 container_of(work, struct nvme_ctrl, scan_work);
4213
4214         /* No tagset on a live ctrl means IO queues could not created */
4215         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4216                 return;
4217
4218         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4219                 dev_info(ctrl->device, "rescanning namespaces.\n");
4220                 nvme_clear_changed_ns_log(ctrl);
4221         }
4222
4223         mutex_lock(&ctrl->scan_lock);
4224         if (nvme_scan_ns_list(ctrl) != 0)
4225                 nvme_scan_ns_sequential(ctrl);
4226         mutex_unlock(&ctrl->scan_lock);
4227
4228         down_write(&ctrl->namespaces_rwsem);
4229         list_sort(NULL, &ctrl->namespaces, ns_cmp);
4230         up_write(&ctrl->namespaces_rwsem);
4231 }
4232
4233 /*
4234  * This function iterates the namespace list unlocked to allow recovery from
4235  * controller failure. It is up to the caller to ensure the namespace list is
4236  * not modified by scan work while this function is executing.
4237  */
4238 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4239 {
4240         struct nvme_ns *ns, *next;
4241         LIST_HEAD(ns_list);
4242
4243         /*
4244          * make sure to requeue I/O to all namespaces as these
4245          * might result from the scan itself and must complete
4246          * for the scan_work to make progress
4247          */
4248         nvme_mpath_clear_ctrl_paths(ctrl);
4249
4250         /* prevent racing with ns scanning */
4251         flush_work(&ctrl->scan_work);
4252
4253         /*
4254          * The dead states indicates the controller was not gracefully
4255          * disconnected. In that case, we won't be able to flush any data while
4256          * removing the namespaces' disks; fail all the queues now to avoid
4257          * potentially having to clean up the failed sync later.
4258          */
4259         if (ctrl->state == NVME_CTRL_DEAD)
4260                 nvme_kill_queues(ctrl);
4261
4262         /* this is a no-op when called from the controller reset handler */
4263         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4264
4265         down_write(&ctrl->namespaces_rwsem);
4266         list_splice_init(&ctrl->namespaces, &ns_list);
4267         up_write(&ctrl->namespaces_rwsem);
4268
4269         list_for_each_entry_safe(ns, next, &ns_list, list)
4270                 nvme_ns_remove(ns);
4271 }
4272 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4273
4274 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4275 {
4276         struct nvme_ctrl *ctrl =
4277                 container_of(dev, struct nvme_ctrl, ctrl_device);
4278         struct nvmf_ctrl_options *opts = ctrl->opts;
4279         int ret;
4280
4281         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4282         if (ret)
4283                 return ret;
4284
4285         if (opts) {
4286                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4287                 if (ret)
4288                         return ret;
4289
4290                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4291                                 opts->trsvcid ?: "none");
4292                 if (ret)
4293                         return ret;
4294
4295                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4296                                 opts->host_traddr ?: "none");
4297         }
4298         return ret;
4299 }
4300
4301 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4302 {
4303         char *envp[2] = { NULL, NULL };
4304         u32 aen_result = ctrl->aen_result;
4305
4306         ctrl->aen_result = 0;
4307         if (!aen_result)
4308                 return;
4309
4310         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4311         if (!envp[0])
4312                 return;
4313         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4314         kfree(envp[0]);
4315 }
4316
4317 static void nvme_async_event_work(struct work_struct *work)
4318 {
4319         struct nvme_ctrl *ctrl =
4320                 container_of(work, struct nvme_ctrl, async_event_work);
4321
4322         nvme_aen_uevent(ctrl);
4323         ctrl->ops->submit_async_event(ctrl);
4324 }
4325
4326 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4327 {
4328
4329         u32 csts;
4330
4331         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4332                 return false;
4333
4334         if (csts == ~0)
4335                 return false;
4336
4337         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4338 }
4339
4340 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4341 {
4342         struct nvme_fw_slot_info_log *log;
4343
4344         log = kmalloc(sizeof(*log), GFP_KERNEL);
4345         if (!log)
4346                 return;
4347
4348         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4349                         log, sizeof(*log), 0))
4350                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4351         kfree(log);
4352 }
4353
4354 static void nvme_fw_act_work(struct work_struct *work)
4355 {
4356         struct nvme_ctrl *ctrl = container_of(work,
4357                                 struct nvme_ctrl, fw_act_work);
4358         unsigned long fw_act_timeout;
4359
4360         if (ctrl->mtfa)
4361                 fw_act_timeout = jiffies +
4362                                 msecs_to_jiffies(ctrl->mtfa * 100);
4363         else
4364                 fw_act_timeout = jiffies +
4365                                 msecs_to_jiffies(admin_timeout * 1000);
4366
4367         nvme_stop_queues(ctrl);
4368         while (nvme_ctrl_pp_status(ctrl)) {
4369                 if (time_after(jiffies, fw_act_timeout)) {
4370                         dev_warn(ctrl->device,
4371                                 "Fw activation timeout, reset controller\n");
4372                         nvme_try_sched_reset(ctrl);
4373                         return;
4374                 }
4375                 msleep(100);
4376         }
4377
4378         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4379                 return;
4380
4381         nvme_start_queues(ctrl);
4382         /* read FW slot information to clear the AER */
4383         nvme_get_fw_slot_info(ctrl);
4384 }
4385
4386 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4387 {
4388         u32 aer_notice_type = (result & 0xff00) >> 8;
4389
4390         trace_nvme_async_event(ctrl, aer_notice_type);
4391
4392         switch (aer_notice_type) {
4393         case NVME_AER_NOTICE_NS_CHANGED:
4394                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4395                 nvme_queue_scan(ctrl);
4396                 break;
4397         case NVME_AER_NOTICE_FW_ACT_STARTING:
4398                 /*
4399                  * We are (ab)using the RESETTING state to prevent subsequent
4400                  * recovery actions from interfering with the controller's
4401                  * firmware activation.
4402                  */
4403                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4404                         queue_work(nvme_wq, &ctrl->fw_act_work);
4405                 break;
4406 #ifdef CONFIG_NVME_MULTIPATH
4407         case NVME_AER_NOTICE_ANA:
4408                 if (!ctrl->ana_log_buf)
4409                         break;
4410                 queue_work(nvme_wq, &ctrl->ana_work);
4411                 break;
4412 #endif
4413         case NVME_AER_NOTICE_DISC_CHANGED:
4414                 ctrl->aen_result = result;
4415                 break;
4416         default:
4417                 dev_warn(ctrl->device, "async event result %08x\n", result);
4418         }
4419 }
4420
4421 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4422                 volatile union nvme_result *res)
4423 {
4424         u32 result = le32_to_cpu(res->u32);
4425         u32 aer_type = result & 0x07;
4426
4427         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4428                 return;
4429
4430         switch (aer_type) {
4431         case NVME_AER_NOTICE:
4432                 nvme_handle_aen_notice(ctrl, result);
4433                 break;
4434         case NVME_AER_ERROR:
4435         case NVME_AER_SMART:
4436         case NVME_AER_CSS:
4437         case NVME_AER_VS:
4438                 trace_nvme_async_event(ctrl, aer_type);
4439                 ctrl->aen_result = result;
4440                 break;
4441         default:
4442                 break;
4443         }
4444         queue_work(nvme_wq, &ctrl->async_event_work);
4445 }
4446 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4447
4448 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4449 {
4450         nvme_mpath_stop(ctrl);
4451         nvme_stop_keep_alive(ctrl);
4452         nvme_stop_failfast_work(ctrl);
4453         flush_work(&ctrl->async_event_work);
4454         cancel_work_sync(&ctrl->fw_act_work);
4455 }
4456 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4457
4458 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4459 {
4460         nvme_start_keep_alive(ctrl);
4461
4462         nvme_enable_aen(ctrl);
4463
4464         if (ctrl->queue_count > 1) {
4465                 nvme_queue_scan(ctrl);
4466                 nvme_start_queues(ctrl);
4467         }
4468 }
4469 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4470
4471 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4472 {
4473         nvme_hwmon_exit(ctrl);
4474         nvme_fault_inject_fini(&ctrl->fault_inject);
4475         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4476         cdev_device_del(&ctrl->cdev, ctrl->device);
4477         nvme_put_ctrl(ctrl);
4478 }
4479 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4480
4481 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4482 {
4483         struct nvme_effects_log *cel;
4484         unsigned long i;
4485
4486         xa_for_each(&ctrl->cels, i, cel) {
4487                 xa_erase(&ctrl->cels, i);
4488                 kfree(cel);
4489         }
4490
4491         xa_destroy(&ctrl->cels);
4492 }
4493
4494 static void nvme_free_ctrl(struct device *dev)
4495 {
4496         struct nvme_ctrl *ctrl =
4497                 container_of(dev, struct nvme_ctrl, ctrl_device);
4498         struct nvme_subsystem *subsys = ctrl->subsys;
4499
4500         if (!subsys || ctrl->instance != subsys->instance)
4501                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4502
4503         nvme_free_cels(ctrl);
4504         nvme_mpath_uninit(ctrl);
4505         __free_page(ctrl->discard_page);
4506
4507         if (subsys) {
4508                 mutex_lock(&nvme_subsystems_lock);
4509                 list_del(&ctrl->subsys_entry);
4510                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4511                 mutex_unlock(&nvme_subsystems_lock);
4512         }
4513
4514         ctrl->ops->free_ctrl(ctrl);
4515
4516         if (subsys)
4517                 nvme_put_subsystem(subsys);
4518 }
4519
4520 /*
4521  * Initialize a NVMe controller structures.  This needs to be called during
4522  * earliest initialization so that we have the initialized structured around
4523  * during probing.
4524  */
4525 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4526                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4527 {
4528         int ret;
4529
4530         ctrl->state = NVME_CTRL_NEW;
4531         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4532         spin_lock_init(&ctrl->lock);
4533         mutex_init(&ctrl->scan_lock);
4534         INIT_LIST_HEAD(&ctrl->namespaces);
4535         xa_init(&ctrl->cels);
4536         init_rwsem(&ctrl->namespaces_rwsem);
4537         ctrl->dev = dev;
4538         ctrl->ops = ops;
4539         ctrl->quirks = quirks;
4540         ctrl->numa_node = NUMA_NO_NODE;
4541         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4542         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4543         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4544         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4545         init_waitqueue_head(&ctrl->state_wq);
4546
4547         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4548         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4549         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4550         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4551
4552         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4553                         PAGE_SIZE);
4554         ctrl->discard_page = alloc_page(GFP_KERNEL);
4555         if (!ctrl->discard_page) {
4556                 ret = -ENOMEM;
4557                 goto out;
4558         }
4559
4560         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4561         if (ret < 0)
4562                 goto out;
4563         ctrl->instance = ret;
4564
4565         device_initialize(&ctrl->ctrl_device);
4566         ctrl->device = &ctrl->ctrl_device;
4567         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4568                         ctrl->instance);
4569         ctrl->device->class = nvme_class;
4570         ctrl->device->parent = ctrl->dev;
4571         ctrl->device->groups = nvme_dev_attr_groups;
4572         ctrl->device->release = nvme_free_ctrl;
4573         dev_set_drvdata(ctrl->device, ctrl);
4574         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4575         if (ret)
4576                 goto out_release_instance;
4577
4578         nvme_get_ctrl(ctrl);
4579         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4580         ctrl->cdev.owner = ops->module;
4581         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4582         if (ret)
4583                 goto out_free_name;
4584
4585         /*
4586          * Initialize latency tolerance controls.  The sysfs files won't
4587          * be visible to userspace unless the device actually supports APST.
4588          */
4589         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4590         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4591                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4592
4593         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4594
4595         return 0;
4596 out_free_name:
4597         nvme_put_ctrl(ctrl);
4598         kfree_const(ctrl->device->kobj.name);
4599 out_release_instance:
4600         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4601 out:
4602         if (ctrl->discard_page)
4603                 __free_page(ctrl->discard_page);
4604         return ret;
4605 }
4606 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4607
4608 /**
4609  * nvme_kill_queues(): Ends all namespace queues
4610  * @ctrl: the dead controller that needs to end
4611  *
4612  * Call this function when the driver determines it is unable to get the
4613  * controller in a state capable of servicing IO.
4614  */
4615 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4616 {
4617         struct nvme_ns *ns;
4618
4619         down_read(&ctrl->namespaces_rwsem);
4620
4621         /* Forcibly unquiesce queues to avoid blocking dispatch */
4622         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4623                 blk_mq_unquiesce_queue(ctrl->admin_q);
4624
4625         list_for_each_entry(ns, &ctrl->namespaces, list)
4626                 nvme_set_queue_dying(ns);
4627
4628         up_read(&ctrl->namespaces_rwsem);
4629 }
4630 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4631
4632 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4633 {
4634         struct nvme_ns *ns;
4635
4636         down_read(&ctrl->namespaces_rwsem);
4637         list_for_each_entry(ns, &ctrl->namespaces, list)
4638                 blk_mq_unfreeze_queue(ns->queue);
4639         up_read(&ctrl->namespaces_rwsem);
4640 }
4641 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4642
4643 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4644 {
4645         struct nvme_ns *ns;
4646
4647         down_read(&ctrl->namespaces_rwsem);
4648         list_for_each_entry(ns, &ctrl->namespaces, list) {
4649                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4650                 if (timeout <= 0)
4651                         break;
4652         }
4653         up_read(&ctrl->namespaces_rwsem);
4654         return timeout;
4655 }
4656 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4657
4658 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4659 {
4660         struct nvme_ns *ns;
4661
4662         down_read(&ctrl->namespaces_rwsem);
4663         list_for_each_entry(ns, &ctrl->namespaces, list)
4664                 blk_mq_freeze_queue_wait(ns->queue);
4665         up_read(&ctrl->namespaces_rwsem);
4666 }
4667 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4668
4669 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4670 {
4671         struct nvme_ns *ns;
4672
4673         down_read(&ctrl->namespaces_rwsem);
4674         list_for_each_entry(ns, &ctrl->namespaces, list)
4675                 blk_freeze_queue_start(ns->queue);
4676         up_read(&ctrl->namespaces_rwsem);
4677 }
4678 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4679
4680 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4681 {
4682         struct nvme_ns *ns;
4683
4684         down_read(&ctrl->namespaces_rwsem);
4685         list_for_each_entry(ns, &ctrl->namespaces, list)
4686                 blk_mq_quiesce_queue(ns->queue);
4687         up_read(&ctrl->namespaces_rwsem);
4688 }
4689 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4690
4691 void nvme_start_queues(struct nvme_ctrl *ctrl)
4692 {
4693         struct nvme_ns *ns;
4694
4695         down_read(&ctrl->namespaces_rwsem);
4696         list_for_each_entry(ns, &ctrl->namespaces, list)
4697                 blk_mq_unquiesce_queue(ns->queue);
4698         up_read(&ctrl->namespaces_rwsem);
4699 }
4700 EXPORT_SYMBOL_GPL(nvme_start_queues);
4701
4702 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4703 {
4704         struct nvme_ns *ns;
4705
4706         down_read(&ctrl->namespaces_rwsem);
4707         list_for_each_entry(ns, &ctrl->namespaces, list)
4708                 blk_sync_queue(ns->queue);
4709         up_read(&ctrl->namespaces_rwsem);
4710 }
4711 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4712
4713 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4714 {
4715         nvme_sync_io_queues(ctrl);
4716         if (ctrl->admin_q)
4717                 blk_sync_queue(ctrl->admin_q);
4718 }
4719 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4720
4721 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4722 {
4723         if (file->f_op != &nvme_dev_fops)
4724                 return NULL;
4725         return file->private_data;
4726 }
4727 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4728
4729 /*
4730  * Check we didn't inadvertently grow the command structure sizes:
4731  */
4732 static inline void _nvme_check_size(void)
4733 {
4734         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4735         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4736         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4737         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4738         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4739         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4740         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4741         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4742         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4743         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4744         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4745         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4746         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4747         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4748         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4749         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4750         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4751         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4752         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4753 }
4754
4755
4756 static int __init nvme_core_init(void)
4757 {
4758         int result = -ENOMEM;
4759
4760         _nvme_check_size();
4761
4762         nvme_wq = alloc_workqueue("nvme-wq",
4763                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4764         if (!nvme_wq)
4765                 goto out;
4766
4767         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4768                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4769         if (!nvme_reset_wq)
4770                 goto destroy_wq;
4771
4772         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4773                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4774         if (!nvme_delete_wq)
4775                 goto destroy_reset_wq;
4776
4777         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4778                         NVME_MINORS, "nvme");
4779         if (result < 0)
4780                 goto destroy_delete_wq;
4781
4782         nvme_class = class_create(THIS_MODULE, "nvme");
4783         if (IS_ERR(nvme_class)) {
4784                 result = PTR_ERR(nvme_class);
4785                 goto unregister_chrdev;
4786         }
4787         nvme_class->dev_uevent = nvme_class_uevent;
4788
4789         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4790         if (IS_ERR(nvme_subsys_class)) {
4791                 result = PTR_ERR(nvme_subsys_class);
4792                 goto destroy_class;
4793         }
4794         return 0;
4795
4796 destroy_class:
4797         class_destroy(nvme_class);
4798 unregister_chrdev:
4799         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4800 destroy_delete_wq:
4801         destroy_workqueue(nvme_delete_wq);
4802 destroy_reset_wq:
4803         destroy_workqueue(nvme_reset_wq);
4804 destroy_wq:
4805         destroy_workqueue(nvme_wq);
4806 out:
4807         return result;
4808 }
4809
4810 static void __exit nvme_core_exit(void)
4811 {
4812         class_destroy(nvme_subsys_class);
4813         class_destroy(nvme_class);
4814         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4815         destroy_workqueue(nvme_delete_wq);
4816         destroy_workqueue(nvme_reset_wq);
4817         destroy_workqueue(nvme_wq);
4818         ida_destroy(&nvme_instance_ida);
4819 }
4820
4821 MODULE_LICENSE("GPL");
4822 MODULE_VERSION("1.0");
4823 module_init(nvme_core_init);
4824 module_exit(nvme_core_exit);